Univac_II_Marketing_Manual_Jun57 Univac II Marketing Manual Jun57
Univac_II_Marketing_Manual_Jun57 manual pdf -FilePursuit
Univac_II_Marketing_Manual_Jun57 Univac_II_Marketing_Manual_Jun57
User Manual: Univac_II_Marketing_Manual_Jun57
Open the PDF directly: View PDF 
.
Page Count: 282
| Download | |
| Open PDF In Browser | View PDF |
TAB LEO F CON TEN T S
PART I (MC 581)
Section
Subject Matter
0000
Introduction and General Comments on EDPM Systems
0900
Equipment Price Lists
1000
Magnetic Tape Units and Control Units
2000
Random Access Storage Units and Control Units
PART II (MC 582)
Section
Subject Matter
3000
Input Devices, Except Tapes
4000
Output Devices, Except Tapes
5000
Central Processing Units and Systems Considerations -Large-Scale EDPMs
6000
Central Processing Units and Systems Considerations -Small-Scale EDPMs
7000
Auxiliary Equipments
8000
Data Processing Characteristics with EDPMs
8500
Specific Applications Information
9000
Miscellaneous
9900
Index
0000
0000
INTRODUCTION
Our common job--whether we be salesmen, sales support personnel, research
engineers, product development engineers, logical designers, applications
specialists, programmers, product planners, or adrninistrators--is selling
UNIVAC data processing systemse This manual is designed to help you, as
the salesmen and sales support staff ultimately responsible for getting
the order, do a more effective job ~:Its sole purpose is to present factual information on our own and competitive EDPMs, together with discussions and evaluations of equipment performance, which will enable us to
meet, with the greatest effectiveness possible, the proposals of other
manufacturers, and to promote the entire UNIVAC line as not only the first,
but also the most efficient, of the electronic data processing equipments.
0
0001
ORGANIZATION OF THIS MANUAL
The manual is organized into sections dealing with the various types of
components of an EDPM system--the input devices, the storage devices,
the central processor, etc.. With this arrangement, it does not include
all comments necessary to compare a Univac II with a 705, for example, in
one place; the compa'risons are included in the applicable major section
which discusses each type of component a The reason for this choice is to
eliminate the duplication which would otherwise occur in comparing Univac
II with the IBM 705, or the DATAmatic, or the BIZMAC; or the File-Computer
with the IBM 650 or the Datatrone Everything said about an IBM tape unit,
for instance, applies to any"equipment with which it may be associated ..
Consequently, each type of component is discussed and evaluated in the
section pertaining to its general category; the section devoted to central processing units includes not only those factors which are peculiar
to the "main frame", but also ties together the many system ramifications
possible into one coherent whole and provides an overall evaluation of
complete systemso
A straightforward system of decimal index numbering is employed, with
major breakdowns indicated in the covering table of contents of this manual.. The basic number is four digits in length, anyone of which may be
sub-indexed with up to three digits to the right of the decimal point ..
Pages are not numbered sequentially, but each has~ in the upper right
corner~ a·large 5.iLndex number which identifies the particular page and its
proper sequenceo Thus, page 88880132 follows page 8888013, which in turn
follows 8888010
The first, or thousands digit, indicates the major section of the manual ..
The second, or hundreds 9 digit is assigned to major manufacturers, although
this cannot be carried through all equipment makerse However, a second
digit tlO" pertains to Remington Rand Univac equipments;a ttltt to IBM
equipment and a tl2ft to DATAmatic components
Specific sales pointers are
always included in sections with a units digi t tt9 tt --i Cle 0' ttll09 .. tt
<>
,avery effort has been made to cIDoss-reference the manual when a knowledge
of material contained elsewhere is necessary or desirable for a complete
understanding of a topic" To avoid the necessity of numerous' correction
sheets as information is added, deleted, or changed, such cross-references
ooe].
0001
(Continued)
usually refer to basic section numbers. Thus it should be understood
that a reference to "section l120 K includes the range Ul120" through
1'1129," and a reference to "1122" includes all breakdowns wi thin that
index number.
Section 9900 contains a subject matter index of the contents of this
manual. An additional index to sections dealing with specific equipments or components is contained in the price lists in section 0900-0999;
The entries for each type number or trade name under which equipments
and components are known include the manual section in which it is discussed.
The discussion of each equipment or component is broken into three parts:
(1) A brief summary of the basic characteristics of the unit under consideration; (2) a discussion and evaluation of the operating features;
and (3) sales points pertaining to it. The second part is usually the
longest and is developed in sufficient detail to permit a ready understanding of the method of operation of a component and an evaluation of
its performance in comparison with the corresponding UNIVAC series equipment. The sales points are largely based upon discussions contained in
this second part.
General comments on EDPM processing and information on specific applications is contained in section 8000. This is intended to serve as background information which will be useful in assisting potential customers
to gain a proper perspective on the relative performances of various
equipments through an understanding of the basic business file processing
techniques which are possible with EDPMs of varying general characteristics.
The development of this manual assumes a reasonable familiarity with the
characteristics and operations of equipments of Remington Rand UNIVAC.
All competitive equipments are explained in a degree of detail sufficient
to permit understanding their method of operation; it is intended that
these explanations be comprehensive enough to eliminate the need for
handbooks or manuals describing their method of operation-- they may not
always be available, and frequently are not exhaustive or complete in
their explanationsD
0002
INTERCHANGE OF INFORMATION
Although every effort has been made to verify the accuracy of all factual
and analytical information contained in this manual, it is inevitable that
some errors or omissions occur. Any which may be noted should be brought
to the immediate attention of the Sales Support Staff, in order that they
may be corrected.
Similarly, the development of timely and accurate comparative analyses
of equipments or components requires the constant availability of CUrrent and complete information on new developments and new areas of
0002
0002
(Continued)
emphasis by competitive firms. Information not present in the manual
should be forwarded to the Sales Support Staff as soon as possible,
in order that a timely evaluation can be made andtfrebuttal U data
prepared. This includes not only new equipments or peripheral units,
but also new or unusual equipment configurations and processing techniques which may be proposed. Only by knowing of these developments
'can they be taken into consideration in improving the effectiveness of
our UNIVAC sales efforts and in evaluating the impact of new equipments
or techniques on the marketability of our own future product plans. The
success and usefulness of a manual of this nature is directly correlated to
the degree to which the joint efforts of all of us in Remington Rand Univac
are co-ordinated to one common end--the widest possible distribution of
the accumulated knowledge of everyone of us. This manual is intended
to provide--at least in part--amedium for accomplishing this.
0900
0900
PRICES OF ELECTRONIC DATA PROCESSING EQUIPMENTS
This section contains the currently-quoted prices for the EDPM equipments and components of leading manufacturers; these are grouped by
manufacturer and are in type number or component name sequence, as
applicable. Each section includes all EDPMs, or components, which
have been or are being made, as well as those which have been announced for future delivery; inclusion does not mean that they are
available for current delivery. It also includes type numbers of
equipments or components which have been assigned, although the unit
may never have been made and may not be planned for production. In
general, these are noted.
Where components can be used in more than one basic computing system,
notation is made of the possible systems into which it may be incorporated. The notation "Dba." in the column headed It'Status, 1t means that
the unit is now out of production and no further orders can be taken;
most of them are of course still in use.
A question mark preceding the price indicates lack of certainty or contradictory information; these will be corrected as soon as specific and
accurate cost data are known.
Some manufacturers assign names or type numbers to such units as power
supplies, memory units, etc., although their cost is usually included
in that for the assembly with which they are used. These are included
in the list for the sake of completeness; in all instances, they are
noted to indicate the equipment or component with which used.
0110
0110
IBM ELECTRONIC DATA PROCESSING EQUIPMENTS
The chart on the next page ties together the equipment relationships
of the EDPMs offered by the International Business Machines Corporation.
This charts includes all functional components available or announced,
but omits many type numbers which are always associated and used with
one or more of the equipments shown: these compris~·, such components as
power supplies, memory storage elements, operator's consoles, etc. A
complete list of type numbers, with references to the basic EDPM systems
with which used, is included in 0910.
Original 3/1/57
0910
0910
PRICES OF IBM EDPM AND PUNCH CARD EQUIPMENTS
Section 0910.1 lists the prices and basic EDPM systems of all known type
numbers assigned to this class of equipmento Where available, both rental
and purchase prices are included. The listing is in type number sequence,
and includes a reference to the section of this manual in which a discussion of the type appears.
The basic processing systems are listed in capital letters; all units not
so identified are component or auxiliary parts of one or more of the
major 'systems. Up to the present time, the assignment of type numbers by
IBM has been in certain broad classes of components, which permits a partial identification of the nature of the equipment by a knowledge of its
type number. This is not, however, universal. The basic groups are:
300 Series:
600
n
700 - 709
710 - 730
731 - 739
741 - 749
750
t59
760 - 779
780 - 789 :
Equipments in the RAMAC 305 series
Equipments in the Type 650 systems
Analytical and Control Units, ~ large-scale systems
Card readers, card punches, printers and magnetic tape
units
Central processor storage devices--drums and magnetic
cores
Power supply and power distribution units
Tape control units
Special purpose tape control units
OperatorVs consoles
In the past several months, IBM has adopted a general policy of prIcIng
equipments by type numbers; for example, the Analytical and Control Unit
of the 705 is priced at $11,800 monthly (Modell), although it requires
a Type 745 Power Supply Unit (at $1,200) and a Type 782 OperatorVs Console (at $1,000) to be operativeo Insofar as practicable, this list
groups associated components under the major one and prices the total;
the individual type numbers are separately enumerated in the "Description"
column, and also appear in proper sequence. An asterisk following a price
indicates 0 unit which is a fundamental part of some other type number.
An "x" keying a type number to a major system indicates a component that
can operate under central processor control; it may also be capable of
periphery or off-line operationo A "y" key indicates a unit that can be
used only in off-line operation in that systemo
The discussions of major systems, sections 5000 and 6000, contain a complete description of the various system configurations which are possible
and also include examples of what may be considered typical installations
of such major systemso
The prices of some punch card equipments manufactured by IBM are listed
in 0910.20 This is not exhaustiveo It is included for reference purposes in arriving at costs of punch card equipment necessary in an EDPM
installationo
Original 2/15/57
MONTHLY COSTS OF IBM EDPM EQUIPMENT
TYPE MOOE~ g r l
g
ro
N o;;:t t.O
0 0 0
r-- r--
Q'\
r-- g
t.O
0
("f)
x
305
I
323
340
350
1
355
370
380
475
533
537
x
x
x
x
I
x
x
x
x
x
x
x
x
I
Ix
x
x
652
x
653
655
x
x
DESCRIPTION
CENTRAL PROCESSING UNIT, RAMAC 305
$1,250
$86,900
Type 305
26,100
Type 340
375
Card Punch
Power Supply Unit for Type 305
Random Access Disc Memory
Random Access Disc Memory
Line ("Stick n) Printer
Operator's Console with Card Reader and InputOutput Typewriter
Early version of Type 370 Printer
Card Read Punch Unit (dual feed)
Card Read Punch Unit (single feed)
MAGNETIC DRUM PROCESSOR
1,000 Word Capacity
2,000 Word Capacity
Type 655
650
I
STATUS
I
$1,900
2,400
1,100
$100,700
127,000
58,300
COST
PURCHASE
ltENTAL
SECTION
~19625
6150
$113 9 000
225
375*
650
975
300
14,600
26,100*
48,880
72,200
19,800
6150
2110
2115
4140
400
27,800
6150
--
---
-550
700
3170
3175
I
3,250
I 3,750
6100
6100
I
Type 650 Optional Features (no type numbers) :
Floating Decimal Point Feature
Indexing Registers (3)
Printer Input-Output Device for use with Type 407
tabulator (numeric only)
Alphabetic Device for use with Type 407
"11" and "12" Punch Device for use with
Types 407 and 533
Special Character Device (all including "11 "
and tt12") for use with Types 407 and 533
Magnetic Tape and Random Access Control
Unit
Record Storage Unit
Power Supply Unit for Type 650 with one InputOutput Synchronizer
? 300*
? 240*
6111
6112
500*
325*
6113
6113
25*
6113
o
\0
.....
100*
9751,950
2,425
1,100*1
I
55,600-
I
58,300*
o
6113
1120
I
"
.....
1125
6100
I
I
Revised 3/1/57
Monthly Costs of IBM EDPM Equipment (contVd)
TYPE
MODEL
701
702
C\I -=::t 1.0
0
0
M
\,()
t- t-
1.0 0
(j\
STATUS
Obs.,
x
Obs.
Obs.
Obs.
2
x
x
x
x
x
705
1
x
705
706
708
709
x
Obs.
Obs.,
x
1
x
x
x
711
711
712
714
1
x
x
x
DESCRIPTION
("I)
x
x
703
704
1.0
ggg0
x
x
YYx Y x Y
Obs.
Additional Synchronizers (each)
CENTRAL PROCESSING UNIT
CENTRAL PROCESSING UNIT:
10,000 Characters Electrostatic Storage
20,000 Characters Core Storage
See Type 770
CENTRAL PROCESSING UNIT:
4,096-Word Storage (Type 737 Modell
8,192-Word Storage (Type 737 Models 1 and 2)
32,768-Word Storage (Type 738)
$523,800
$9,700
Type 704
Floating Point Feature (additional cost)
CENTRAL PROCESSING UNIT, 20,000 CHARACTERS
$613,600
Type 705-1 $11,800
1,200
Type 745
1,000
Type 782
CENTRAL PROCESSING UNIT, 40,000 CHARACTERS
$749,600
Type 705-2 $14,300
1,200
Type 745
1,000
Type 782
Electrostatic Storage Unit for Type 701
Modified Type 704; superseded by Type 709
CENTRAL PROCESSING UNIT:
4,096 Word Storage (Type 737 Model 3)
8,192 Word Storage (Type 737 Models 3 and 4)
32~768 Word Storage (Type 738)
$600,000
Type 709-1 $10,000
1,400
72~800
Type 741-2
96,000
1,600
or Type 741"",3
1,300
67~600
Type 746
Punch Card Reader (125/minute)
Punch Card Reader (250!minute)
Punch Card Reader (C-T Converter)
$36,800
$750
Type 712
300
Type 756
Punch Card Reader (C-T Converter)
COST
RENTAL PURCHASE
$ 300*
5120
5120
8~650
10,150
=-=
SECTION
6100
5110
=-
-=--
5130
5130
5130
15~800
20,800
29,700
14~000
5130
5140
16,500
5140
4~000*
1,600* $ 28,100*
-......"
--
16,900
20,100
32,900
1,141,400
200
800
1,050
52,000
948~400
1~03,600
5110
-
5150
5150
5150
3110
3110
3120
o
\,()
.
I-'
o
I-'
2,400
3120
Revised 3/1/57
Monthly Costs of IBM EDPM Equipment (cont'd)
nTI.
715
716
717
719
720
721
722
726
727
729
730
731
732
733
733
734
736
736
737
737
737
737
738
740
741
741
741
742
MQ!2§1.
0
.-I
(\j
~
If)
0' J,()
0 0 0 0 0
0
\0 l"- l"- l"- l"- l"- (V)
If)
1
x
x
x
Y Yx Y x Y
1
x y
x rY
x
x
Y y x rt
x
Y
Y
x
x
x x Pc
1
Y
x
x ~
x y
x y
x
x y
Obs.
x y
x
x y
Obs.
x
1
2
x
x
~
~
x
1
2
1
2
3
4
1
1
1
2
3
STATUS
x
x
k
x
~
x
x
x
x
x
x
x
x
x
x
x
P<
x
DESCRIPTION
COST
RENTAL PURCHASE
SEC- .
TION
$1,500
$97,500
Type 714
$ 900
Type 759
Output Recording Typewriter
78,050 4110
Alphabetic Printer (150 lines/minute)
~l ,200 $
1,800
4120
Alphabetic Printer (150 lines/minute)
Type 717
$1,200
$73~950
600
Type 757
74,200 4130
1,400
Alphabetic Printer (lOO~minute, 60 characters)
74,200 4130
1,400
Alphabetic Printer (500 minute, 120 characters)
39,000 4510
600
Punch Card Recorder (Card Punch) (100/minute)
4520
1,050
Card Punch (T-C Converter)
$750
$44,400
Type 722
300
Type 758
42,100
850
Magnetic Tape Unit
29,800 1100
550
Magnetic Tape Unit
35,750 1105
650
Magnetic Tape Unit
149,900 4130
2,775
Alphabetic Printer (lOOO/minute, 120 characters)
87,450
1,400
Magnetic Drum and Power Unit
$1,100
$57,200
Type 731
300
30,250
Power Supply
121,900 2520
2,300
Magnetic Drum and Power Unit
167,400 2510
3,100
Magnetic Drum (8,192 36-bit words)
189,000 2510
3,500
Magnetic Drum (faster transfer than Model 1)
2520
2,800
Magnetic Drum (60,000 characters)
5130
-Power Unit No.1; rental included with CPU
-57,200* 5150
1,100*
Power Supply Unit
5130
6,100*
Magnetic Core Storage (4,096 words)
5130
5,000*
Magnetic Core Storage (Next 4,096 words)
208,000* 5150
4,000*
Magnetic Core Storage (4,096 words)
193,000* 5150
3,200*
Magnetic Core Storage (Next 4,096 words)
20,000* 1,040,000* 5150
Magnetic Core Storage (32,768 words)
162,000 4910
2,700
Cathode Ray Tube Output Recorder
5130
Power Supply Unit; rental included with CPU
72,800* 5150
1,400*
Power Supply Unit
"96,000* 5150
1,600*
Power Supply Unit
2520
-for
Type
732;
rental
included
with
it
Power Supply Unit
---
--
-
Revised 371 57
(
"..
<
..
.0
Monthly Costs of IBM EDPM Equipment (cont'd)
IXE.§.
MODEL
0
L()
\.0
0
2g
r-- r-- ~
r-- r-- r--
,......j
743
x
1
2
3
x
Obso
x
x
x
x Y x ~
x
x
x
x
x
x
x
x Y
Y
Y x Y x Ii
y
x y x Ii
x Y x IY
Y
Y
1
Obso
x~
I
1
2
pc
P<
Y
775
777
780
781
782
795
838
STATUS
~
1
I
Obs.
Dc y Dc Ii
P<
pc
pc
I
Obso
x
Obs.
Obso
P<
DESCRIPTION
(Y)
x
771
772
774
797
1.0
0
x
744
745
746
746
746
747
752
753
754
755
756
757
758
759
760
766
766
770
N
0
Power Supply Unit; rental included with CPU
Power Supply Unit; rental included with 650
Power Supply Unit
Power Distribution Unit
Power Distribution Unit
Power Distribution Unit
Power Supply Unit for Type 774
Tape Control Unit
Tape Control Unit
Tape Control Unit
Tape Control Unit
Control Unit for Type 712
Control Unit for Type 717
Control Unit for Type 722
Control Unit for Type 714
Record and Storage Unit
Data Synchronizer
Data Synchronizer
Special Sorter, originally known as Type 703, built
for National Security Agency; no commercial plans
Data lacking
Data lacking
Tape Data Selector
$1,700
$102,000
Type 774
500
Type 747
Record Storage Unit; announced but never delivered
Tape Record Coordinator
7-inch Cathode Ray Display Tube
Console and Typewriter; rental included with Type 702
Console and Typewriter
Special purpose machine built at Northrop Aircraft.
no commercial plans
Core storage version of Type 795; IBM built four
but has no commercial plans
Inquiry Keyboard in RAMAC 650
_
COST
PURCHASE
RENTAL
....
210,000
210,000
SECTION
5120
6100
5140
5110
5150
5150
1170
1130
1135
1140
1145
3120
4120
4520
3120
1150
1160
1160
-=-
=-
-=---
---
~,200*
~,300*
~,300*
~~300*
67,600*
67,600*
500*
550
R,500
28,600
67~600*
104,000
108,000
~,OOO
~,800
300*
600*
300*
900*
~,850
~,500
~,500
=-=-
~,280
83,200
37,100
570
~,200
~,850
~,OOO
1170
I
111,000
156,000
150
8~700
-=-
-
~,OOO*
---
-
5140
-=-
-
--
=--
175
--=>
1180
4910
I
10,000
--
7110
Revised 3/1/57
o
\0
I--'
.o
I--'
0910 .. 2
091002
PRICES OF IBM PUNCH CARD EQUIPMENT
Following is an incomplete list of IBM punch card equipments, with monthly
rentals and sales prices.. It does include nearly all of the units which
may be required as auxiliariesto an EDPM installation. Prices shown are
probably for basic equipments without any of the optional devices which
may be added.. Some units are indicated with two or three different cost
figures; these may represent different models or versions of the same basic equipment, but are not otherwise explained in the partial information
from which this list has been compiledo More accurate costs will be provided when available.
Type
Number
Cost
Rental
Purchase
Equipment
010
016
024
026
Key Punch
Duplicating Key Punch
Alphabetic Key Punch
Printing Key Punch
027
028
043
046
Punch Card Proof Machine
Printing Punch Card Proof Machine
Tape-Controlled Card Punch
Tape-to-Card Punch
t1
"
n
047
056
065
066
067
075
077
080
t1
$
"
t1
Tape-to-Card Printing Punch
Verifier
Data Transceiver Card Unit
Printing Data Transceiver Card Unit
Telegraph Signal Unit (Transceiver)
Card Counting Sorter
Numeric Collator
Sorter
t1
082
083
086
089
101
402
Sorter
Sorter
Sorter
Alphabetic Collator
Statistical Sorter
Alphabetic Electric Accounting Machine
403
407
Alphabetic Electric Accounting Machine
Alphabetic Electric Accounting Machine
419
513
514
Numeric Electric Accounting Machine
Reproducing Punch, Mark-Sense
Reproducing Punch, Mark-Sense
"
"
"
"
"
"
"
"
"
519
523
n
"
n
End-Printing Reproducing Punch, M/S
n
"
"
f1
Gang-Summary Punch
If
n
n
t1
n
It
10
40
35
60
55
145
160
70
140
135
160
50
90
110
85
70
100
40
25
55
110
100
220
500
420
395
470
835
920
425
40
125
110
135
120
85
85
80
$
700
1,950
1,750
3,700
3 11 500
8,000
9,500
3,000
8,400
8,100
9,600
2,400
4,700
6,450
4,950
4,550
6,600
2,600
2,565
3,400
6,550
6,900
14,300
30,000
27,950
27,450
31,150
54,250
59,850
26,550
1,350
7,500
6,600
7,650
6,850
5,150
5,550
5,200
Original 3/1/57
0910.2
(2)
0910.2
(Continued)
Type
Number
526
528
549
552
557
602
Printing Summary Punch
n
It
"
Accumulating Reproducer
Ticket Converter
Alphabetic Interpreter
Alphabetic Interpreter
Calculating Punch
It
604
605
607
608
803
824
826
916
923
Cost
Purchase
Rental
Equipment
"
Calculating Punch
"
It
Calculating Punch (part of CPC)
Calculating Punch
Transistorized Calculating Punch
Bank Proof Machine
Typewriter Card Punch
Typewriter Card Punch
Bill Feed
Tape-Controlled Carriage
$
100
95
440
250
75
165
245
220
400
510
615
650
1,600
135
95
115
50
30
$
5,900
5,700
26,400
12,500
4,650
9,400
15,900
26,000
26,000
37,100
40,000
35,800
10,125
4,700
5,900
1,950
1,950
Original 3/1/57
0940
0940
PRICES OF DATAMATIC CORPORATION EDPM EQUIPMENTS
The DATAmatic Corporation is wholly-owned by Minneapolis-Honeywell.
Description
1000
1052
1054
1056
1058
1090
1100
1150
1170
1200
1300
Central Processing Unit, including
Model 1090 Console and facilities
for controlling up to 8 tape units
Auxiliary Central Processor; controls
up to 2 additional tape units
Same; controls up to 4 additional
tape units
Same; controls up to 6 additional
tape units
Same; controls up to 8 additional
tape units
Operator~s Console; cost included with
Model 1000
Magnetic File (Tape) Units
Magnetic File Reference Unit, including typewriter
Magnetic File Switching Unit
Input Converter (Card-to-Tape Converter); includes card reader but
not magnetic tape unit
Output Converter (Tape-to-Card or
Tape-to-Printer)0 Does not include
tape unit, card punch or printer.
Output Converter and high-speed
printer. Includes converter and
printer, but not tape unit
_ _~C;:;.;0:or:s~t_
Purchase Rental
Section
$985,000
$21,500
5410
100,000
2,200
5410
175,000
3,900
5410
250,000
5,500
5410
295,000
6,600
5410
60,000
1,350
1400
50,000
9,600
1,075
190
7410
1410
185,000
3,325
3410
100,000
1,800
4410
215,000
4,300
4420
Original 6/1/57
0950
0950
PRICES OF
BURRO~SCORPORATION
EDPM EQUIPMENTS
Prices of equipments manufactured by Burroughs Corporation and its .Electrodata Division are included. It is understood that all future development
of EDPMs will be under the Electrodata Division.
Cost
Purchas.§. Rental
Des cJ:..i..Qti on
EIOI
ElOl
ElOl
EIOI
204
205
360
403
405
406
409
446
454
458
500
506
543
544
560
E101 Computer, with keyboard input
semi-ganged pl.'lntel.', 100-word
$
magnetic drum, two pinboard sets
Additional Pinboards
Additional 120 wordsoof drum capacity
Paper Tape Reader (20 digits/second)
Paper Tape Punch
32,500
$
Section
850
3,000
3,500
3,500
Basic Datatron Central Processor, including magnetic drum and arithmetic
135,000
unit
Magnetic Drum. part of Model 204 and
cost included with it
Floating Decimal Ari thme'ti.c Unit for
21,100
use with Model 204 Datatxon
7,050
Console, for use with Model 204
Consollett.e (Remote Control), for use .
1,980
with Model 204
Console , with photoelectl'ic pe:rforated
tape reader and tape punch
Console, with photoelectric perforated
tape reader (no tape punch)
Photoelectric Reader (used with Models
2,960
406 and 409)
2,040
Paper. Tape Perfo:rato:r., used with 406
2,980
Paper Tape PeTforator~ high-speed
4,560
Typew:ri ter (Flexowriter) Control Unit
?
Paper Tape Preparation Unit
3,135
Flexow:r:t.ter, Program-Cont:rolled
18,625
Punch Card Converter Unit
Cardat:ron:
31,000
Control Unit and Power Supply
22.,500
Input Station
26,300
Output Station
25 ,000
Magnetic Tape Control Upit
12,000
Datareader (Magnetic Tape Unit)
25,000
Datafile (Multiple Magnetic Tape Bin)
I
80
95
3610
4710
3,900
6520
725
230
6520
6520
70
6520
451
6520
362
6520
95
3620
4720
4720
7620
7610
7620
3450
95
61
89
137
?
95
567
770
560
660
750
375
825
lBA punch card equipments usable with Model 204 Datatron:
500 series sumrna.:ry punches (input/output)
089 Collator
400 series Tabulators
I
6510
25
60-455
190-220
205-800
Original 6/1/57
3460
3465
4610
1520
1510
2510
0960
PRICES OF LOGISTICS RESEARCH, INC., EDPM EQUIPMENT
0960
These prices include ·only those of components used with the ALWAC 800
EDPM equipment.
;
Model
801
DESCRIPTION
COST
Purchase Rental
882
Central Processing Unit (includes
1,000 words of core memory)
Magnetic Drum Buffer Unit
Magnetic Drum
Punched Card Buffer Unit
Magnetic Tape Buffer Unit
Magnetic Tape Transport Unit
Paper Tape and Keyboard Buffer Unit
Electric Keyboard
High-Speed Paper Tape Reader
High-Speed Paper Tape Punch
Additional 1,000-Word'Magnetic Core
Storage, for Type 801 (each)
Address Modifier Unit for Type 801
885
Floating Point Arithmetic Control Unit
810
815
820
840
845
860
865
866
867
881
Section
$325,000
$11,580
5510
28,000
34,000
53,000
25,000
18,000
15,000
5,000
1,500
1,500
950
1,150
1,530
850
610
510
170
85
40
2850
2850
4850
1510
1520
3510
7510
3520
4860
75,000
6,500
2,550
220
5510
5510
23,000
740
5510
Original 6/1/57
1100
1100
IBM TYPE 727 TAPE UNIT
The Type 727 is the standard tape unit used with the Types 650, 702,
704 and 705 EDPMs and computer systems, both for on-line and off-line
operationso A variety of control units, listed in 1102, is provided
for various purposeso The Type 727 always requires a control unit to
be usedo
1100 1 Monthly rental:
0
1101
$550.
TAPE UNIT AND MAGNETIC TAPE CHARACTERISTICS
Essential Specifications are listed below.
Magnetic Tape and Reel
Tape base:
Tape width:
Tape length:
Tape reel diameter (standard):
Tape reel weight (empty):
Tape weight:
Plastic container diameter:
Plastic container thickness:
Tape cost, including reel:
Type 727 Tape Unit
Number of recording channels:
Character recording density:
Tape speed: forward
rewind
Character transfer rate
Tape start time:
Tape stop time:
Tape reversal time:
Bad spot detection:
Detecting beginning of tape:
Detecting end of tape:
Method of recording:
Length of tape record:
Rewind time:
Check bit:
Setting of tape unit addresses:
Accetate (scheduled to be replaced to
mylar)
1"
"2
200, 600, 1200 or 2400 feet (standard)
lot"
12 ounces
1 ounce per 100 feet
11 9/16 inches
1 7/16 inches
2400 feet of tape -1200 feet of tape -600 feet of tape --
(acetate base)
$50
$32
$22
7, each group of 7 bits recorded in
parallel
200 characters per inch
75 inches per second
500 inches per second, except last 400
feet (75 inches/second)
15,000 per second
10 milliseconds
604 milliseconds maximum
60 milliseconds
None: tape must be perfect
Photoelectrical sensing of reflective
spot
Same, or tape mark
Non-return-to-zero; every valid character
must have a binary "1"
Variable, subject to restrictions imposed by CPU characteristics
Maximum of about 102 minutes for any
length of tape
Normally eveno Odd when recording
binary information on 704
Rotary switch on each unit
Revised 1/31/57
1102
1102
CONTROL UNITS AND USES OF TYPE 727
The various configurations of control units and basic computer
systems or periphery operations with which the 727 may be associated
are listed below. Refer to writeups on the various control units or
periphery functions for more detailed information.
Type 650 EDPM:
Types 702/705 EDPM:
Type 705 EDPM:
Type 704 EDPM:
652
754
760
774
777
755
Control
Control
Control
Control
Control
Control
Unit,
Unit,
Unit,
Unit,
Unit,
Unit~
maximum
maximum
maximum
maximum
maximum
maximum
of
of
of
of
of
of
6 - 727s
10 - 727s
2 - 727s
1 - 727
8 - 727s
10- 727s
Periphery Operations:
Card-to-Tape Conversion:
Tape-to-Card Conversion:
Printing:
Punch Card Operations:
712
714
727
727
727
727
727
727
727
727
727
-
756
759
758
757
760
760
760
774
774
774
774
-
727
727
722
717
719)
720) One or two tape units
730) may be connected
402 )
402A) Tabulators
407 )
519 Reproduci,ng Punch
1103
1103
MAGNETIC TAPE FORMATS WITH THE TYPE 727
Although the 727 is used with all current IBM computers, the appearance
of information stored on tape is a function of the particular equipment
which created it. Tapes are nominally interchangeable between and among
systems, provided the restrictions or requirements of each are met. Use
of this tape unit with the Tape 701 EDPM is not discussed.
1103.1
Common Features of Tape Information Storage
Regardless of the initiating computer, or tapes resulting from· C-T conversion, all uses of Type 727 tapes have some common features in the
recording of information and tape handling. All specifications of 1101
are applicable to any use. The basic tape layout is shown below, the
numbers attached to the arrows indicating the sub-paragraph in this
section in which the feature is discussed.
r1
013 .14
1103.11
t
012
013
t
r. r
•12
1-'
.. 14
.. ll
.13
.15 •. 11
Reflective Spot: Adhesive aluminum-coated strips, are placed several
feet from each end of the tape to indicate the physical beginning and
ending points for recording information. The spots are about i·' x I·'
in size and are placed in the positions indicated, allowing enough
footage to provide for the vacuum loop controls and two or three turns
around the hub of the tape reel. A minimum of about six feet is required on each end, but 10-15 are more normally used. The spots can
be moved quite readily by the user. In both reading and writing, the
tape is positioned just past the beginning reflective spot as a part
of the loading operation; this is automatically done by the 727 through
a photoelectric cell sensing device. It is impossible to read or write
on the end of the tape past this spot. In writing only, the end-of-tape
reflective spot is similarly sensed and a program-testable signal is generated. Unlike the beginning-of-tape spot, however, it is possible to
write beyond the end-of-tape spot. Sensing of the end-of-tape spot occurs only during tape writing; the circuitry is inactive during reading.
(See 1105.15).
1103.12 Records are written as a variable number of characters, partly dependent
upon characteristics of the EDPM being used. Recording is in seven
channels, all seven bits being handled in parallel. The record consists
of 6-bit information codes plus a check bit.
1103.13
Check Character. Each record is followed by a check character, automatically created by the equipment in all input/output operations. In
both reading and writing operations, the number of binary fils" on each
of the seven channels is counted and the check character created to make
the total number of ·'lst., including the check character, an even number.
Physically, it is separated from the last information character by a space
equivalent to three character positions (0.015 11 ) . These three "blank"
spaces are identical to a recording of seven binary zeros and thus are
1103.13
1103.13
(Continued)
identical to the inter-record gap (1103.14); consequently, every
information character must contain at least one ·'1. it The sale
use of the check character is to verify the accuracy of reading
information previously recorded on tapes. During writing, the
character is created and placed on tape at the end of the record;
during reading, the character is created while the record is being
read and the recreated check character compared with the original.
The entire process is automatic and requires no attention by the
programmer.
1103014
Inter-Record Gap. The physical space on tape separating successive
records, is 3/4 ft , from the check character of the previous record
to the first information character of the next. The variable lengths
of information records have no effect on the length of the gap.
1103.15
Tape Marko The last record written on a tape is followed by a single
special character (0 00 1111), called the tape mark, which is placed
there by a special instruction; each system has such an instruction.
In writing a full reel,the end-of-tape reflective spot normally is
passed over while writing a record, and the tape mark thus is closer
to the physical end of the tape than the spot. It may be many feet
past it, because sensing the reflective spot does not disable the writing
circuits, and several records may be placed on the tape after the spot
is reached. On partially-filled reels of tape, the tape mark is placed
after the last information record, and is separated from it by an interrecord gap. In reading a tape, sensing of the tape mark generates a
program-testable signal indicating the end of recorded information on
the reel; this is the only method (other than special program-generated
record counts) of recognizing this conditiono
TAPE RECORD FORMATS OF IBM EDPMS
The only variations of tape recording occur in the records themselves;
the reflective spots, check characters and inter-record gaps are common
to all equipments5 Figure 110302 shows the types of record formats
possible with various equipments, together with the data in memory
which represents the record. The check character is not shown; however,
it should be remembered that it follows every tape record, regardless
of generating equipment.
The formats contained in Figure 110302 are discussed in more detail
beginning in 1103.3
1IO'S~3
1103.3
TAPE FORMAT WITH TYPE 650 EDPM
Two basic formats are possible, one for strictly numeric and one for
alphanumeric or mixed information.
1103.31
Numeric Information. Numeric data, contained in memory as 7-bit biquinary coded digits in words of 10 digits plus sign, are carried digitfor-digit on tape in standard 702/705 character codes, not in 650 codes.
The conversion in either direction is performed automatically in the
Type 652 Tape Control Unit. On the basis of read/write times contained
in 650 instruction manuals, it appears that the sign digit is not written as a separate character, but is included as a zone code in the units
position of each 10-digit word on tape; this is consistent with the 702/
705 method of handling signs. Changing of the sign digit to a zone
code, and vice versa, is done automatically as a part of input/output
operations.
Any number of words up to 60 maximum (600 digits plus 60 signs) may be
placed on tape as one record. Numeric tape records used in 650 operations consist, then, of anything from 10 to 600 tape characters in
multiples of 10'; partial words cannot be written or read. Both reading
and writing instructions must specify the exact number of words involved;
the exact number specified are always written out and, in reading, if the
number of words in the actual record is greater or less than that specified,
an error indication results.
1103.32
Alphabetic or Mixed Information.. Internally in the 650, alphanumeric
information requires two biquinary coded digits for each character; one
word can be ten digits or five alphanumeric characters; the sign position
is disregarded in an alphanumeric word. Each word is an individual entitYi strictly numeric words can be mixed with alphanumeric oneso
Alphanumeric or mixed information can be placed on tape only in multiples
of ten words: 10, 20, 30, 40, 50 or 60 (maximum).. An alphanumeric word
in 650 memory is converted automatically to five 702/705 character codes
when written on tape, and conversely when reado Alphanumeric and strictly
numeric words may be mixed, but anyone word must be one or the other.
In 650 memory, the last (tenth) word of each ten-word group is a £Qlltrol
designating the status of the previous nine words.. An u8" in a
control word digit position indicates that the corresponding word of the
group is alphanumeric; a "9," that it is purely numeric. The tenth digit of
the control word is always a "9,1t because it designates the numeric status
of the control word itself. Any or all of the rema1n1ng nine information
words may be alphanumeric. (Control words are obviously not required for
strictly numeric information).
~
In writing on tape, the control word precedes the nine information words,
which (presumably) follow in sequence, and is used to control the method
of writing out the nine information words. A word with a control digit
"8'1' is written as five (alphabetic) 702/705 character codes; with control
digit "9 ft as ten (numeric) codes, identical with 1103.31. This use of
the control word to govern the method of writing out the ten-word group
is automatic. In the following example, the nine information words are
1103.32
(Continued)
1103.32 -
identi fied as "Atf for alphanumeric and "Nit for numeric; all
digits of the control word are shown.
650 memory:
A
N
_ _N__A_.J ~~8989..1
N ~_~~,__A
Information Words
Codes on tape:
No. characters:
8899898989
A
ANN
10
5
5
\..-----y---1 L.....__
Control Word
10
10
Control Word
A
5
NAN
10
.-y-----
5
10
A
5J
-
Information Words
The maximum number of alphanumeric characters is a ten-word group is
45 (nine words of five each). The maximum number of alphabetic characters in a single tape record of 60 memory words is 270 (45 x 6).
This requires 330 tape characters, because the control words are
written out in numeric form.
In reading from tape, the control word of each ten-word group is read
first and governs the automatic reconversion of the 702/705 character
codes into the correct 650 codes. The position of the control word is
changed to the tenth word of each group.
As is the case with numeric information, read and write instructions
must specify the exact number of words (which must be multiples of 10)
involved; otherwise, errors result.
1103.4
1103.4
TAPE FORMAT WITH TYPE 702 EDPM
Information character codes used in the 702 are identical with the tape
characters. Tape records may be any length from one character up to one
less than memory capacity (9999 or 19999)0 The writing of a tape record
(with or without a buffer in the control unit) proceeds character-bycharacter beginning at the addressed position and continuing until a
special character code, called a record mark (1 01 1010), denoted by
tt=l=, tt is reached in memory. The record mark is not placed on tape, but
sensing its presence initiates the end of the writing operation (with,
of course, writing of the check character and creation of the interrecord gap) ..
In reading, recognition of the inter-record gap (actually, the three
blank positions preceding the check character) terminates the read
operationo A record mark is automatically inserted in memory in the
position following the last information character read in.. Figure
1103 .. 2 will clarify the relationsh~p of the information in memory with
the corresponding tape record.
By a special provision in the address of the ttwri te tl instruction, it
is possible to write out all of memory, record marks included, beginning at any addressed character position and ending in the next-to-last
character position (memory address 9998 or 19998), as one tape record.
On reading in such a record, record marks in the tape record are treated
as any other valid character code. However, at the end of the operation
a record mark is automatically inserted in memory following the last
information character
If the record was a "memory dump,11 beginning at
address 0000, the record mark would be inserted in 9999 or 19999; this
is the reason the last character position is not written out when this
special "write" provision is used.
0
It is important to note that a record mark in memory always terminates
a tlwrite tl operation on the 702 (with the exception noted above), but it
does not appear on the tape record itself.. If a record mark is included
as a character code in a tape record (as in a memory dump or on a tape
prepared on the 705), it is treated as any other valid character; it
does not end a reading operation. A record mark is still placed in
memory follbwing the last character read in; this is the one generated
by sensing the inter-record gap.
1103.5
TAPE FORMAT WITH TYPE 704 EDPM
1103.5
Two types of formats are possible with the 704, one for pure
binary information and one for character-coded (702/705 codes).
1103.51
Binary Tape Records. The 36-bit binary internal word is represented
on tape as six characters of six bits, plus check bit, each. Any
number of words may be included in one tape record, but the total
number of tape characters normally must be an integral multiple of
6--i.e., full computer words. Unlike the other IBM systems, there
is no theoretical limit to the length of a tape record, except the
physical length of the tape itself.
There is one other difference in this method of recording. Because
long strings of binary zeros are possible in data, an even check bit
used in the other systems, could easily result in several successive
tape characters of nothing but binary zeros. These characters would
be identical with the '''blank'' spaces indicating the end of each tape
record; thus, in a reading operation, their presence would be taken
as indicating the end of a record. To overcome this difficulty, an
odd check bit, assuring a binary ttl" in every information character,
is used in binary recording. The check character, however, is based
on the even channel bit count used in the other equipmentse
1103.52
Character-Coded Tape Records. Although binary in its arithmetic operations, the 704 can contain information in coded form internally; these
can be written out as standard 702/705 character codes. It is not necessary to write them out in this fashion, however--they can be handled as
if they were binary data. The character codes used in the 704 differ
from those of the 702/705 in three respects:
(1) ~cimal zeros are 00 0000 in the 704 and 00 1010 (plus check
bit) in the 702/705; they must be converted.
(2) All alphabetic zone codes used in the 704 differ from those of
the 702/705 and must be converted
(3) An even,not odd, check bit must be inserted on tape characters.
The 36-bit, 6-character coded word in the 704 is written out as six tape
characters, in 702/705 codes, each with an added even check bit. Successive groups of six bits are converted automatically to the proper
702/705 character codes. Like binary recording, each tape record normally
must be an integral multiple of six characters, but can be of almost
unlimited length.
1103.53
Record Lengths Not a Multiple of Six Characters. The 704 writes tapes
only in full words--that is, 6 groups of six binary digits or six coded
characters. However, it can read a record which is not an even multiple
of six tape characters, provided its exact length is known. The requirement of known length is necessary because the method of transferring the last partial group of bits is different from transferring a
full grouY). Evidently, tapes so read must have been prepared on a 650,
702 or 705.
1103.6
110306
TAPE FORMAT WITH TYPE 705 EDPM
Essentially, tape records used with the 705 follow the principle of
those used with the 702 (1103.4), with one important addition: The
record mark (:F) of the 702 is treated in input/output operations as
any other valid character code, and a new code, called the group ~
(0 11 1111), designated by "~", serves exactly the same function as
does the record mark in the 702. The group mark is not a valid character in the 702, but because it does not appear on tape, a 705-created
tape can be used as input to a 702 (or vice versa)o
1103.61
Record. In IBM terminology, a record is a group of characters, variable
in length, terminated in memory by a record mark. On a 702~created tape,
the record mark is replaced by the inter-record gap; on a 705 tape, it
appears as a valid character code and does not terminate the writing
o~eration.
1103.62 Grouped Record. In the 705, several records, separated by record marks,
may be treated as a grouped record, terminated in memory by a group
mark and on tape by an inter~record gap. In the 705, the record marks
terminating each (except the last) record also appear on tape. Grouped
records cannot be written on the 702, but they can be read; in this case,
a record mark (not a group mark) is inserted following the last character
in memoryo The group mark is peculiar to the 7050
1103.7
1103.7
TAPE FORMAT WITH TYPE 709 EDPM
Two types of formats are possible with the 7090
1103.71 Binary Tape Records.
1103.51.
Recording is identical with the Type 704.
1103.72 Character-Coded Tape Records.
See 1103.52.
See
Recording is identical with the Type 704.
1103.73 Record Lengths Not a Multiple of Six Characters. Like the 704, the 709
writes tape records, in either method of recording, only as an integral
multiple of six characters. Unlike the 704, it can read a record which
is not an even multiple of six characters without advance knowledge of
the length; the 704 must know the exact length to read such records. The
partial group of characters (five or less) is automatically assembled in
the high~order (left) end of the 709 word before transmission to memory.
Such tapes must of necessity have been prepared on a Type 650, 702 or
705.
~iginal 2/15/57
1104
1104
ERROR AND SIGNAL INDICATORS PERTAINING TO TYPE 727 USAGE
All systems using Type 727 tape units incorporate checking of tape status
and accuracy of read/write operations. There are several basic conditions to be taken into account. Each of them can cause an automatic stop
of the EDPM, but some of them can be set for program interrogation and
control.
1104.1
Tape Unit Not Readyo This condition arises when no tape unit has been
set to the address selected, or when a valid tape unit is in what IBM
calls a "not ready" status--tape rewinding, no tape reel mounted, or
mounted but not yet at the load (starting) point.
1104.11
Type 650 error indication not known.
1104.12
Types 702/705 stop automatically, with the "I/O No Response" neon on the
operator's console lighted.
1104.13
Type 704 indication not certain, but apparently the computer stops, with
the "Read/Write Check Indicator t9 turned on.
1104.2
Read/Write Character Code Errors. In writing, each character code is
given a redundancy bit test (except in the 704, where a check bit is
merely added to each six memory bits), and the check character is created.
In reading, each character is tested and also the check character is recreated and verified against that on the tape& These operations are automatic. Error indications vary.
1104.21
Type 650. The "Branch No Tape Signal tt conditional transfer instruction
permits a conditional transfer to be made if either a tapeerror or endof-file (tape) condition existso This can be followed by a "Branch No
End of File" instruction, which distinguishes between the two conditions
and causes a program transfer to the appropriate sub-routine.
1104.22
Types 702/705. An error in either reading or writing sets the "Read/
Write Check Indicator." This is an addressable indicator (address 0902)
which, by the setting of a console switch, may be handled either as an
automatic machine stop or a program-controlled correction. In the latter
case, a conditional transfer instruction on the status of the "Read/Write
Check Indicator" provides a branching if an error has been made; otherwise, the transfer address is ignored and the normal program continues.
1104.23
Tvpe 705. A character code error turns on the uTape Check Indicator," on
the console, and makes available a program-testable signal. This indicator
does not directly cause a machine stoP? but the conditional transfer instruction testing it can be branched to an automatic stop if desired. In
addition to the validity test, this indicator is turned on if a "read" instruction in binary mode is used for a character-coded tape, or vice versa.
1104.24
Type 705. The "Read/Write Check Indicator" may be tested by the "Transfer Any" instruction of the 705, as well as by direct interrogation.
This instruction is not included in the 702.
1104.3
End-of-Tape. On writing, sensing of the reflective spot signifying approach of the physical end of the tape generates a testable signal. This
condition never results in an automatic stopo
Presumably, the computer stops.
1104.31
1104.31
Type 650. The 11Branch End of File" instruc'tion causes a conditional
transfer to be made when the reflective spot is sensed.
1104.32 lypes 702b05. An "Input..{)utput Indicator" is turned on and can be
used in a conditional transfer instruction. A separate indicator
is provided for each tape unit and selecting a unit automatically
makes the indicator available.
1104.33
Type 704. The "Tape Check Indicator l1 is turned on and may be programtested. Apparently, the existence of an end-of-tape condition can be
distinguished from a writing error only by time of occurrence.
1104.34
Type 705. The "Input-Output Indicator U may be tested by the''Transfer
AN'lt1 instruction.
1104.4
End-of File. On reading, the tape mark rather than the reflective spot
signifies the end of onformation on a tape. The indications are the
same as end-of-tape: 1104.3; the only difference is the means used to
turn them on. The programmer, of course, knows which tapes are used
for reading and which for writing, and can readily provide for the
proper tape-switching subroutines to be called up by the conditional
transfer instruction.
1104.5
Incomplete Word on Type 704. An attempt made to read, in the normal
manner, a partial group of less than six characters turns on the "Tape
Check Indicator." This condition can~ri·sa~ only on the final characters
of records on tapes prepared on a 650, 702, or 705, where records are not
necessarily in multiples of six characters. Refer to 1103053 for 704
input records not a multiple of six characters in length.
1104.6
Incorrect Number Qf Wo;ds on Type 650. Both read and write instructions on the 650 specify the number of words involved. Because it always writes the number of words specified, no error results in this
operation. On reading, if a tape record has fewer or more words than
indicated in the instruction, the machine stops with an error indication on the console. This type of error cannot be program-corrected.
1105
1105
IBM TYPE 729 TAPE UNIT
The Type 729 is the tape unit used with the Type 709 EDPM in central
processing operations. Essentially, it is a modified version of the
Type 727 (section'llOO) and produces magnetic tapes which are completely
compatible with those of that unit.
Purchase:
Rental
1105.1
~:
$35,750
650 monthly
1106
TAPE UNIT AND MAGNETIC TAPE CHARACTERISTICS
Essential specifications are listed below.
Magnetic Tape and Reel
Identical with 'those used wi th Type 727; see 110L.
729 Tape Unit
All specifications for the Type 727 Tape Unit, 1101, are applicable.
In addition, one new characteristic is added:
~
Verification of Tape Writing:
Characters are read immediately after
being written for check bit validity; this is accomplished by having
a set of "read" heads immediately
following the "write" heads.
Functionally, this is the only feature of the Type 729 which distinguishes
it from the Type 7270
1107
CONTROL UNITS AND USES OF THE TYPE 729
In on-line operations, the Type 729 is used only with the 709 EDPM and
not with the other major systems of IBM. The actual linkage is accomplished through a Type 755 Tape Control Unit, which provides power and
some control circuitry, connected to one channel of a ~ype 766 Data Synchronizer. Each channel may have up to eight Type 729s connected through
one Type 7550 One Type 709 may have up to three Data Synchronizers, each
wi th two channels; if eight tape units are connected to each channel,q,\
(through a Type 755 for each channel), a maximum of 48 Type 729s may 'be
associated with a Type 709 CPU.
With the exception of its use with the Type 774, it is not known definitely if the
Type 729 can be used in lieu of a Type 727 in the periphery (off-line) operations
listed in 1102. Because it is functionally identical with the Type 727, there
appears to be no reason why it is not interchangeable in periphery operations
and this assumption will be made until contradictory information is received.
Sales points on the Type 729 begin in section 1109.50.
Original 2/15/57
1108
1108
PURPOSE OF THE READ-CHECK FEATURE OF THE TYPE 729 TAPE UNIT
The functional characteristic which distinguishes the Type 729 from
the Type 727 is the inclusion of the flread-check tt or verification of
tape writing in the 729. The emphasized purpose of this change is to
provide facilities for proving the accuracy of tape recording immediately
after the writing. The unemphasized reason is probably a good deal 'more
important ..
1108.1
Causes of Tape Reading Errors. Several causes can result in failure to
read a tape record. From the technical standpoint, the list can be rather
long, but from the purely functional viewpoint, the following are the more '
important in the IBM method of recording on and reading tape.:
1108 11
Incorrect Spacing of Pulses!)
fOllow each other at specific
This is discussed in 1109007;
the 729 has no effect on tape
1108.12
~G
0
Reading IBM tapes requires that the pulses
time interval s, plus or minus a toler.ance.
the inclusion of the read-check feature in
reading errors arising from this cause.
This also is discussed non-technically in 1109.07 .. Again, the 729
does not correct read failure axising from skew difficulties.
1108.13 Faulty Pulse R!t,cordinq on Good T~o Even though ~ parity bit test is
made on the pulses sent to th~ write heads, it does not follow that the
resulting magnetization of the tape is good" The energizing of the heads
may be such that the :resulting pulse is weaker or stronger than normal,
and reading may give an error because the pulse is too weak or too strong
(saturated, technically); spots on the tape where the oxide coating is
somewhat thinner or thicker than nominal may contribute to this.. On all
magnetic recording devices, there is a certain amount of inherentbackground
"noise," and the info:rmation pulses must be sufficiently strQI9 to rise
above this noise to be discernible.. On th.e other hand, if the information
pulses are too strong, they may Jl9saturate tt the heads and several successive'
pulses may merge into one indistinguishable groups These conditions may
aI'ise even though the tape 15 considered 9' good .. tt
In the Univac tape system, the automatic re-xead features incorporated in
the Uniservos permit varying the ngainllt of the signals; this means that
we can often successfully read ta.pe pulses which are too weak ox too
strong to be picked up at normal gain. The IBM tape units do not have
this feature and always :read a.t a constant llVgain t9 or amplification.. The
read-check feature of the 729 can cope with this condition, at least to
some extento
1108.14 Fl,awson Tape" In use, magnetic tape develops flaws, even though it may
have been 9IperfecttO when it left the factoryo These "flaws" consist of
a vaTiety of different phenomena: Chips in the oxide coating, deposits
of grease, dust ox non-magnetic pa:rticles on the surface, nicks ox dents
in the tape, which prevent the heads from making sufficiently close contact, etc. Some of these conditions can he :removed, but, when present,
neither writing nor :reading can be done with any degree of :reliability.
With Univac tapes, such flaws can be "punched outll'and passed oveI' by the
~bad spotO!! detection circuits in the Uniservos; IBM does not include this
facility in its equipment line.
Original 5/15/57
1108.2
1108.2
Detection of Errors by the Read-Check Heads. Shortly after being written,
the newly-created characters pass under the "read check" heads, which test
the pulses for readability and give the character a parity bit test. The
read heads are physically located about 1/8 ft from the ~ite heads, which is
equivalent to about 25 character positions on tape, or 1.6 ms in timeo
The circuitry performing the validity read is deliberately designed to
work under conditions more adverse than the normal reading circuits. Any
failure to pass the parity bit test is an indication of faulty recording,
and generates a signal which can be tested by an instruction in the main
program of the 709.
Failure to obtain a valid parity bit does not indicate the cau'se ; it could
be a pulse too weak or too strong, a flaw on the tape, or a temporary
fault in the checking circuitry itselfo The same signal is generated, regardless of the cause. The normal tape-writing instl''-lction sequence of a
709 program can be expected to include a sub-routine, entered only when a
read-check error occurs. This routine will attempt to write the tape record
again, and repeat the cycle three or four times. If continued errorS occur,
one of two cond! tions most likely exists,: ( 1) There is a permanent failure in the tape writing circuits themselves; or (2) there isa flaw on the
tape.
Isolating the first cause can be accomplished rather easily by typing out
a short messag,e on the opexator' s console whenever a read-check error occUrso If there isa permanent failure in the writing circuits, there
should be repeated failuxes of this nat\lI.'e, and continued type-outs are a
signal to the opexatox to call in the maintenance engineers.
What course of action isavai1able fox the second case--a flaw on tape?
1108.3
Program-Contxolled ttpassing Oyer" of Bad Spots. Unless there is a failure
in the tape writing circuitry, continued inability to record a xeadable
tape block most likely indicates a flaw on tape. Rather evidently, some
means must be provided for '"passing over tl this bad spot; there is no point
in writing a block on a piece of tape known to be bad.
In the 709 system, bad spots on tape can be signalled and detected under
complete program control. In other words, a program can be written to
accomplish the same thing we do with Uniservos with the bad spot '"punch
outs." This facility is available because there are two methods of recording on tape, and two write and two read instructions which specify
the method usedo One of these uses an even check bit; the other, an odd.
(See 110305) .. Thel'ead instruction must specify the type of recording
and an errol' signal, which can be tested by a program instxuction, isgenerated whenever the parity bit does not agree with that specified& Although IBM has not announced the specific techniques it will advocate, the
general approach of this pJ:'ogrammed bad spotellmination i~ quite simple.
1108031 Bad S ot Elimination With Cha:racte:r-Coded Recoxding_ SUppose a tape is
record.ed with 702 705 character codes, which have an even check hit. If,
in the writing ope:ration~ a continued xead-check failure indicates the
probability of a flaw on tape, this technique pxovides an easy means of
detecting the condition on a subsequent :reading ope:ration. The :routine
Original 5/15/57
1108.31
1108.31
(Continued)
attempting to re-write the tape block will stop after three or four tries,
and move ina sequence of instructions which place on the tape several
inches of "hash tt recording having an odd check bit (binary recording). In
reading this tape, the ttread tt instruction will specify an even check bit;
reach!ng this "hash" block, with the incorrect type, generates a readerror signal which is tested as a normal part of a 709 tape-reading instruction sequence. Because this type of error cannot be distinguished
from other types (for example, simple failure to read validly), some additional checking is necessary. This can be done by making the "hash tt
block consist of bit combinations which repeat a specific character; for
instance, it may be nothing but "$" symbols repeated for 1,000 characters
(five inches on tape). If the input-error sub-routine tests several of
the character positions for a "$" character code, positive identification
of a bad spot, rather than some other type of failure, can be made.
Several positions normally must be tested, beca\lse the mere fact that the
"hash" block is written over an imperfect piece of tape means that not all
character codes will be readable.
1108032 Bad Spot Elimination With Binary Recording. The technique for a tape being recorded in binary, with odd check bits, is just the reverse of the
above. In this case, the tthash" block would be written with an ~ check
bit, each six-bit group repeating a specific bit pattern--for example,
10 1010. This is detected and tested in exactly the same manner as outlined
in 1108.31.
110804
Comments on Programmed Bad Spot Elimination in the 799/709. In announCing
the incorporation of the read-check heads into the 729, IBM has stressed
the fact that they permit verifying the readability and validity of tape
characters as soon as they are written, and to correct improper recording
immed~ately.
The last thing mentioned--and nicely buried at the end of a
paragraph--is this:
t'The 729 tape unit, by constantly checking the quality of the: recording
and the quality of tape surface, achieves more error-free passes, extends the useful life of a reel of tape and increases the number of
times a record may be read. Programming can prevent the recording of
data on defective sections of the tapeo Many marginal tape errors
will be eliminated."
This verbatim quote is an interesting admission from a company that has
required perfect tape with its EDPMs ever since the first one was produced. By the time the first 729 is in use with this feature, IBM 700
series equipments will have been in use for about five years--with perfect tape. The implications of their about-face are obvious.
1108,,5
Comparison of IBM and Univac Methods of Bad Spot Eliminationo Now that
IBM has admitted that the normal tape is t'imperfect ," it· is pertinent" to
compare the corrective methods adopted by them and by Univac. The Univac
technique is to punch out the bad area, a permanent means of eliminating
it from possible recording. The IBM method requires detecting the bad
spot, and pr.ogramminga ,"hash" block, on each tape pass involving writing.
Original 5/15/57
1108.5
1108.5
(Continues!)
The principal differences between the two approaches are two in number:
(1) When the bad spot is detected; and (2) the permanency of the corrective
action. In the 729 method, the ~ad spot is discovered immediately, but is
never permanently flblacked out;fl the corrective routine is performed every
time the tape is used as output. The newly-created tape is, however, good
insofar as bad spots are concerned--no information is written on an imperfect piece of tape. In the Univac method, the bad spot is discovered,
normally, only when the tape is used as an input, either to the CPU or a
periphery equipment. The corrective action, by way of contrast, is done
once and is permanent. It is, however, possible to record information on
imperfect tape and not discover the fact until much later.
Which method is better? The answe~ of course, is the one thqt is the
least wasteful of EDPM time and money_ Rather obviously, it is a function
of the frequency with which new flaws appear on tape and of the total number
which may be present on anyone reel. The determination of what constitutes imperfect tape is itself not definitive. For example, mylarbased tape being inspected for Univac II usage averages about ten bad spots
or punch outs per reel; IBM, using the same type of tape from the same
supplier, is getting perfect reels of comparable length. If nothing else,
this indicates that Univac inspection standards must be more rigid.
However, the user is interested in performance rather than inspection
standards. So long as IBM tapes give satisfactory results, he is not
particularly interested in the rigidity of technical tests. Insofar as
bad spots are concerned, the method which results in the least loss of .
good EDPM time (assuming comparable costs) is the best for the customer.
Which method has the least time loss--Univac or IBM? Unfortunately, the
data to make a specific answer are lacking. The factors to be considered
are nonetheless quite clear.
Because a bad spot on Univac tape is normally discovered only in a subsequent ipput operation, it can be assumed that a rerun of the job creating
the tape is necessary. (It is not often that the~roneous recording can
be corrected manually, although this is sometimes possible). Consequently,
the time loss per bad spot is that necessary to recreate the tape; this
typically will run from 10-15 minutes, ~ncluding machine setup.
In the IBM method, bad spots must be handled on every tape writing operation, at least; the reading instruction sequences can be designed to
eliminate re-reads of flhashfl blocks. The time loss is partly a function
of the number of characters in the record which cannot be written successfully and partly of the number of times it is written before concluding
that the tape is imperfect in the place involved. Suppose that three
tries are made at writing the information block--this is probably a
reasonable number--and further suppose there are 1,000 characters in the
information block. The time required for the three fftries f ' plus that for
the t'hash" block is about 950 ms (seven passes over the tape block at 83.4
ms each, plus six tape reversals at 60 ms each). Thus, about one second
is required for each bad spot, and this is repeated every time a bad spot
occurs.
Original 5/15/57
1108.5
1108.5
(Continued)
If 15 minutes is taken as the time lost by Univac for a newly~discovered
bad 'spot, a total of about 900 could be recognized and program-eliminated
by the 729/709 in the same time. The cost-comparison of the two methods
then reduces essentially to this determination: Assuming fairly constant
and continuous writing of output tapes, What is the cost of Univac main
frame time between discoveries of two new bad spots on tapes versus that
of the 709 for the time requiredto pass about 900 bad spots on tape?
This requires knowledge of several factors not now available: (1) What is
the average number of bad spots passed through Univac output tapes for each
newly-discovered imperfection; (2) what is the average nlUllber of bad spots
per 729/709 tape reel; (3) what are the comparable costs of the main frame
plus associated tape units and controls; (4) what are the relationships
between the character transfer rates of the two equipments; etc. Some of
these are known, or can be determined, but the fil'st two are lacking. A
definitive answer cannot be given until these factors are available.
Original 5/15/57
ll~
1109
SALES POINTS ON TYPE 727 TAPES AND TAPE UNITS
This section includes actual or claimed desirable features of the
727 and methods of rebutting them, as well as its weak points and
how to capitalize on themo There is no significance to the sequence
in which the factors are listedo
1109.01
"BAD Spot~ Detectiono The 727 demands perfect tape; the Uniservo
does not,.
For the 727: Faster read/write times because there are no
!tbad spots U to skip overo
Against the 727: The percentage of Uniservo tape which is punched
out is usually very small and adds little to tape time.
Tapes don't wear out unifomly, but get scattered flaws from
such things as fingernails, abrasive dust particles, etco
With Univac tape, continued use is possible; with the 727,
two shorter pieces result. It should be noted that splicing
is not too effective and seldom practicedo
1109.02
Tape Capacity. IBM tapes in 2400-foot lengths and a 3/4" gap can contain more information than a Univac metallic tape. Typical capacities
are:
Uniservo II
727
Uniservo I
3,600 plus
Number of 720-character records 6,600
2,000
Number of characters
4,750,000 1,440,000 2,600,000 plus
5,000
Number SO-col punch cards
25,;000
5~OOO
Maximum normal capacity
5~0009000 1$1440,000 2,600,000 plus
Mylar tape on Uni$ervo II will give single-reel capacities about the
same as the IBM tape.
For the 727:
Holds almost twice as much information as a Univac II
reel at a comparable cost ($50 a reel against $51 for Univac tape).
Initial tape investment is less than for Univaco
Against the 727~ There is no argument on present capacities, but
mylar tape for Uniservo II will provide equalityo However, note
that tape capacity on the 727 is often a function of file format;
it mayor may not approach the figures given above. On Univac,
tape capacity is independent of formato On the cost aspect, the
point to stress is the cost per time used, not the initial costo
Both are less than a nickel per tape pass and, with Univac tape's
punch-out feature for bad spots, the cost per million characters
stored over a long period may well be less than the 727Q s • At any
event, any cost differential--in favor of the 727 or of Univac-is a fraction of a cent per million characters stored$
1109.03 Rewind Time. The 727 has a high-speed rewind with an approximate
maximum of 102 minutes for any tape rewindo The Uniservo does noto
Rewind is closely related to tape changing~ 1109004$
For the 727: The fast rewind gives several minutes for tape changess
Against the 727: It needs the fast rewind~ because of the slowness
of tape changing. The Uniservo rewind is fast enough for its purposes--namely, to permit tape changing on multi-reel operations
without stopping the computer 0 Part of the higher price of the 727
1109.03
1109.03
(Continued)
is due to the high-speed rewind. As a further note, tapes sometimes break during the sudden acceleration into high-speed rewind
and an insecurely mounted tape reel may fly off the mounting, with
rather disastrous results to the innards of the tape unit. Needless
to say, the tape itself isnVt much good after this treatment.
1109.04 Tape Changing. The Uniservo is considered to have a clean-cut advantage. Therefore, only the steps involved in changing 727 tapes are
listed, with sales points following.
1.
2.
3.
4.
5.
6.
7.
8.
Press "Unload" key and wait several seconds for vacuum columns
to unload and tape drive mechanism to open up.
Open door and manually rewind the last several feet (usually 15
or more) of tape.
Turn pressure band knob several times to loosen reel and remove
ito
Put on new reel, unwinding several feet of tape, and tighten
pressure band knob.
Feed loose end of tape through retracted tape drive mechanism,
into takeup reel and under the holding spring.
Manually wind tape onto takeup reel until reflective spot is well
past the heads.
Close door and press "Load" and "Ready" keys. In 8-15 seconds,
unit is ready.
Set address selector switch, if necessary.
Against the 727.
1.
2.
3.
4.
There are several weak features.
Changing a 727 reel takes It-2 minutes, compared with t minute
for Uniservo. This is one reason the 727 needs a high-speed
rewind.
The reel may not be seated firmly against the back-plate of the
mount. One edge of the tape then rubs against the reel flange;
tape creasing or wrinkling which results doesnVt help reading.
The seating of a 727 tape is strictly manual--uniform pressure
of the operator's hand as he tightens the pressure band knob
with the other hand. Uniservots looking mechanism is not only
faster than the 727°s, but it assures firm seating of the reel.
The locking knob for the pressure band may not be tightened completely. This is a good way to have the reel fly off during highspeed rewind, which has happened. Conversely, some strong-fisted
operator may have given the knob a final extra twist; a weakerfingered tape changer may fight this for quite a while before he
can loosen it. This also happenso
The tape may slip out of the holding spring on the hub of the
takeup reel while the first few turns are being wound manually;
this obviously delays tape changing. Also, getting the end of
the tape inserted under the spring may be quite a feat for an
"all thumbs" operator; it is strictly a "close quarters" job.
1109.05 Multiple Tape Writing. Several 727 tape units may be set to the same
address, permitting simultaneous writing of the same information on more
than one tape. The maximum number differs with the various equipments
with which used. It is a fast way of tape duplication which requires a
separate run on Univac.
1109.05
1109.05 (Continued)
For the.721, This is a desirable feature when more than one copy of
an outpJt tape is necessary and the requirement is known at the time
of initial preparation. For example, one reel can be stored in a
remote location and serve as a spare if the "working copy" is accidentally destroyed; another use is that of a central office sending
duplicate copies of master file information to a number of branches
for local use. When so used, tape d\,lplication in this manner can be
useful.
Against the 727. There is no good rebuttal if a duplicate must of
necessity be prepared. However, to date there is no known use of
duplicate tapes as insurance media" and it is well to ask: "How much
Univac time would be required to do essential duplication?" If the
duplicate is prepared on an IBM equipment simply to avoid a machine
return when one output tape can't be read subsequently, the duplication method becomes a fairly expensive form of operating insurance
not necessary with Univac. (See 1109.07).
1109.06 Tape Reliability_ Occasionally rema:tks' a'te encountered on the relative
merits of plastic versus metailic tape, usually with reference to
the fact that reels or theheaVter metallic tape are shorter in
length and hence contain less'information per reel. Capacity is
discussed in 1109.02. Technically, today's plastic and metallic
tapes are highly reliable, although each has certain technical advantages not possessed by the o,ther.: Univac I uses' metallic-based
tape largely because it was far superior to plastic when a tape choice
had to be made several years agoo Our position is best illustrated by
noting that Uniservo II will :be ~able" tohandle:.either type; as a matter
of fact, at least one user has modified Uniservo I to handle plastic
tape.
Against the 727,. It can use only plastic~based tape, which means that
the more permanent storage qualiti~s of metallic tape are not possible with any IBM system. Furth~tmore, the tape used is acetatebased, not mylar; the latter is being tested on 727s, but so far as
known is not actually in production useo (See; 1109.07 for some comments on the use of mylar with the 727). Uniservos handle either
plastic or metallic; the latter withstands temperatures which reduce plastic to a glop. This can be an important point where ir'~laceable records are to be stored for long periods of time.
1109.07 Readability_ Univac tapes can be read with a much higher degree of accu~
racy than those of the 727. There are two good technical reasons for this;
because our superior tape readability is a major selling advantage, a simple
non-technical explanation of each followso
Skew. By skew is meant the slight twisting which a tape may undergo
as it passes under the read-write headso What it means is that the
pulses don't all reach the heads', when 'reading, at ·the same time; the
pulses from the lagging side of the t~pe reach the heads after the
leading side pulses, and the worse the skew the longer the lag. Now
neither the Uniservo Oor the 727 is always alert for pulses; for technical reasons, the heads are ttopened" momentarfly (a couple microseconds)
and the pulses "sampledQ" A highly-exaggerated illustration will show
1109007
(I)
1109.07 (Continued)
why the Uniservo is a better reader:
r---
)
(
Arrows indicate
0 :
limi ts of read- I
~:~ ,
abili ty on
~I
ei ther side
'>~--- t-I
of heads
. _ J , ;'-1'
ij,
,
1
I 0
b
r",;
:
0
I...J
0,
i
I~··j
I
~; \
---7:
Z,
:Jf
",j'
Cc'-E--
~.-~
-c:
~-.~
j
11
l_
;
,- --71
,
"i'J'~'
rt-·----+
t-1
\,
I
1,1
'~-.:;
"
;".0'
)
.,1,.1 -(
"7
of tape (
motion
~
Here, a ·"1" represents a magnetized spot on tape; a t'{)", an unmagnetized spot. As the tape moves under the heads, the "Is'" cause a magnetic field to increase in intensity to a maximum value and then die
out; reading can occur only when the strength of this magnetic field
is above some minimum, or threshhold, value. Hence, while the entire
rise and fall of the magnetic field may spread over 20 or more microseconds, it may only be strong (big) enough to be usable for a very
few microseconds. Obviously, if a tape is skewed, the peak valuesof
the pulses donVt occur under all heads at the same time, and if the
skew is bad enough, one or two of the magnetized spots may be so far
away from the heads that their magnetic field, at the time of sampling,
is below the threshhold value. And the basic difference between the
Uniservo and the 727 is ~ the pulses are sampledo
H
c'
H
i
I
We use the sprocket pulse--which is a magnetized spot for every character code on tape--to trigger ·the sampling.. Because it is in the
middle of the tape, the leading pulses (when skew is present) are beginning to drop off and the trailing ones are ~till increasing. But
there is still enough signal (unless the skew is quite bad) to read
all of themo If, for example, the heads can recognize pulses from magnetized spots which, timewise, are 5 microseconds away from the heads
the Uniservo can read a tape which is skewed so badly that the leading
side passes under the heads 10 microseconds before the trailing side.
(Needless to say, these times are exaggerated)o
Now for the 727, it has no sprocket pulse, but uses the first pulse
of each character code which comes under the heads to trigger off the
sampling of the entire charactero It may be that the first pulse is
in the middle of a character and no trouble occurso But because the
check bit is one outside channel, the lowest numeric bit is the other,
and the odds are that half of the bits in either of these are "Is,'" it
is rather evident that the presence of skew will cause the sampling to
be triggered off by an outside bit--the beginning of a character-about half the time. And, 'using the exaggerated example of the Uniservo, it is equally evident that a pulse trailing by more than 5
microseconds isn't going to be read.. In other words, if all other
factors were equal, the Uniservo CQuld read a tape skewed twice as
badly as could the 7270 Of course, the other factors aren"t all equal,
1109.07 (Conti~ued)
110,$07
(2)
but it suffices to say that a Uniservo not only reads better, but a
lot better, than a 727.
Time Between Pulses. All tape units write pulses at periodic intervals, timed by an electronic clocko For example, Univac II writes
one character every 50 microseconds, Univac I about every 78 and the
727 about every 66070 In reading tapes, Uniservos pay no attention
to the time interval between successive pulses--they simply look for
a sprocket p\1lse and know that a character is under the heads when one
is sensed. How far apart sprocket pulses may be is immaterial; they
can be 200 to the inch, two feet apart (and frequently are if a bad
spot punch out is reached), or one per tapeo The Uniservo simply reads
until it has counted 720 characters, which can take 51 milliseconds or
51 hours ..
The 727, however, must time its reading as well as its writing operation. This is necessary first, because there is no sprocket pulse and,
second, because the number of characters in a tape record is variable
and lack of a character code is its only signal that a record is finished. It expects a character about every 66,,7 microseconds, and is
only looking ("Sampling") for one that often, with a small deviation
(timewise) on either side. Now suppose that a writing tape unit is
moving a tape somewhat faster than the normal rated speed--which is
a perfectly normal expectancy with a mechanically-actuated operation.
(It happens on Uniservos, for instance),. The characters then are a
bit farther apart than usual, because they are written at fixed time
intervals. Now put this tape on an_other unit moving somewhat slower
than normal--also a normal expectancy. Instead of characters coming
under the heads every 67 microseconds, theyl111 be a bit farther apart,
and take 10ngerG In extreme cases, it can take enough longer to lie
outside the maximum permissible deviation, with the result that the
tape unit assumes itlls reached the end of the record and disconnects
the headso Usually, no permanent error results, because the check
character test f'ails at best, rereading is necessary and~ at worst,
the tape must be changed to another tape unito
For the..l21... Not muchG One less head, with its associated circuitry,
should result in lower cost for the tape unit, and might boost the
727 price if it were included ..
Against the 727.. The higher reliability and readability of Univac
tapes is a major selling point in our systems$ Here are some
points to drive homee
1.
2.
Explain why we read tapes more accurately-- the above is a nontechnical reason0
It is not uncommon practice to write the tape unit number on
727 output labele and to try to place them on the same physical
unit when reading is requiredo This is often impracticable and
usually a nuisance--for example" an output tape which may be
destined for an off-line printing operation.. Univac tapes-regardless of the equipment used--are completely interchangeable within and between installations; 727 tapes cause difficulty in both casesc
1109.07
(3)
1109.07
(Continued)
3
At least one 702 installation rereads every output tape as soon
as it has been created o This means a rewind and a read--7t or
8 minutes with a full reel--during which nothing else is doneo
They do it because experience has taught them that too many
reels canVt be read and their conclusion was that rereading all
tapes took less time than rerunning portions of jobs only when
outputs couldnOt be read subsequentlyo The latter is the conventional Univac practice; Uniservos are reliable enough to make
it cheaper to do a rerun for the occasional tape which does turn
out to be unreadable o
The introduction of mylar tape into the IBM line is already long
delayedo The major difficulty is the fact that mylar stretches
rather readily and the 727's method of reading obviously won't
take much stretch~ which increases the time between pulses o The
Uniservo, of course 9 doesnVt have this trouble o Ultimately, IBM
will probably get it to work, but right now it seems to be all
talk and no performanceo
0
40
50
STRESS UNISERVO RELIABILITYo
1HE 727 IS A WEAK LINK IN THE
ENTIRE IBM COMPUTER LINE AND MANY OF THEIR CLAIMED ADVANTAGES
FALL FLAT BECAUSE OF UNRELIABLE TAPE PERFORMANCE.
1109.08 Fixed Versus Variable Block Lengtho Insofar as tape unit characteristics
are concerned, the major implication of the variable record length of all
IBM equipments is that tapes canVt be read backwards; without knowing the
number of characters in a record it canVt be reassembled properly in
memory.
For the 7270 By writing long records on tape 9 not only is more information stored on one ree1 9 but read-write time is reduced because
there are fewer inter~records gaps and start-stops timeso
Against the 727~ Variable block length is subject to restrictions and
systems considerations; the latter 9 of course 9 are not a function
of tape unit characteristics o
10 Buffer and control unit considerations restrict the size of
tape records 0 In practices 9 702/705 records are limited to ,1,000
characters; 650 records to 600 decimal digits plus 60 signs;
704 records are the only ones not subject to these restrictions.
See Sections 1120 and following for the effects of various types
of tape control units on record length.
2. Inability to read backwards causes sorting and merging operations
with 727s to take more time than if this were possibleo It is
necessary to rewind tapes between each sort-merge, whereas Uniservos read the tapes in both directions s eliminating the flidle fl
time during rewinds
3
Inability to read backwards is another reason for a high-speed
rewind in the 7270
o
0
1109.09
End-of-Tape Indicationo The 727 provides for testing for end of tapes
by the reflective spot (writing) or the tape mark (reading)o In Uniservos, which have no comparable facility, output blocks must be counted
1109.09
1109.09
(Continued)
and tested against a maximum (normal tape capacity) to determine when
a tape is full; input blocks are tested for sentinels. The claim is
made--correctly--that the 727 method permits the test to be made with
fewer instructions and less computing time.
For the 727. IBM points out the ease of testing end-of-tape, both on
reading and writing. It takes only one instruction on the 705, and
no more than two on the 702, for example.
Against the 7270 The theory of the 727 method is fine; the practical
aspects are something else. The basic objection pertains to the
fact that the 727 is completely unchecked in its control functions.
This is more thoroughly discussed in 1109.9, but two effect immediately pertinent are listed here.
1.
2,
For reasons pertaining to lack of certainty on execution of
some tape instructions, some IBM users are numbering all tape
records during writing and checking them during reading. This
is similar to the Univac block-counting method, but the record
numbers are actually placed on tape (adding to read-write time),
a practice not often encountered with Univac. This practice,
of course, makes block or record control fully as time-consuming
as in Univac, offsetting much of the theoretical advantage of
reflective spots and tape marks.
Both reflective spot and tape mark give unchecked indications.
In writing, the spot can fail to give a signal, which results
in ttwriting off the end of the tape. tt This evidently requires
a rerun of the portion of the job creating that tape. Missing
a tape mark in reading is normally not so serious, because anything following it on the tape usually is a partly erased record
which will give a character-check error. Of course, time is
lost in discovering that this type error occurred; the indication of missing the tape mark is exactly the same as failure to
read a record on tape correctly.
1109.10 Character Code Checking. Uniservos are built to give each character
code a redundancy bit check and a 720-character block count; an error
condition results if either fails anyplace during a block. Records on
the 727 also undergo a redundancy bit test, but substitute the check
character test for the character count; with ~ iariable-length record,
something other than a character count is obviously necessary. Insofar
as the adequacy of the test are concerned, both methods are perfectly
sound and there is nothing to indicate one is any better, or any worse,
than the othero However, in certain cases it is possible, with the 727,
to isolate an error to a single character through use of the check bit,
and to determine which channel ~ioeo, bit) is erroneous through a channel error indicator actuated by the check character; such an indicator
is incorporated in some of the control units used with 727s.
For the 727: With the channel error indicator and the check bit for
the character, it is possible to isolate the specific bit in a
specific character which is incorrect (this applies to tape reading).
This permits correcting the error at the operator's console, rather
than rerunning the job which originally created the tape.
1109,10
1109.10 (Contiued)
Against the 727. This is a fine idea, provided the error resulted
from a single bit in a single character of a record which actually
is too weak a magnetic spot to be read successfully (this implies
that it is a binary "I"). This is not usually the case; a bit
missed on one reading has an odds-on chance of being read successfully
the next time, and it's faster to program a tape reread than to attempt manual correction. It is more attractive as a possibility when
a minute tape flaw prevents reading a single bit--a flaw something
like 1/200 by 1/15 inch in size at most. That's a pretty··small flaw.
In fact, in over 15 months of operation of a 702, one user has never
been able to correct a record reading error in this fashion. It is
a theoretically attractive sales point of relatively minor importance,
and of practically no use (if any) in actual operation.
1109.11 Method of assigning Tape Unit Addresses. The logical (program) address
of a 727 unit is set by a rotary selector switch on each unit. The address of Uniservos is determined by wiring in a small plugboard on the
main frame. Relative advantages and disadvantages of the two methods
are probably not of any great importance.
For the 727: If a tape unit malfunctions during a run and the tape
can be changed to another unit, the address change can be made by
the tape changer in the 727 method. With Uniservos, a plugboard
wire--not physically located next to the tape unit--must be changed.
Against the 727: The plugboard method used with Uniservos not only
assigns positively the units required for a given run, but also
automatically disconnects those not needed. With the 727, units not
required for a machine run must be set to an address not used in the
program. In addition, there is some chance of error because the tape
handler may not change the address of a unit to its new logical address, or may set it incorrectly. However, proper housekeeping
programs should test the status of all units needed in a machine
run and any incorrect settings can be corrected before the main
body of the program begins. At most, inconvenience and some processing delay results.
1109.12 Tape Storage Requirements. Plastic-based magnetic tape--particularly
acetate--is best stored in a temperature and humidity-controlled area;
this is not mandatory for metallic tapes. In general, it seems most
EDPM users are providing good storage facilities for tapes, usually
air-condi tioned. Metallic tape does seem to withstand shipment--par-'
ticularly by air--somewhat better than plastic.
For the 727. Adequate storage is normally provided anyway.
Against the 727. As used in IBM equipments, the tape must be at
normal computer room temperature before use. If stored in a
place below recommended temperature limits, it must be removed
and allowed to t'warm up" to normal temperature before use.
This is not necessary with Univac metallic tapes.
1109013
1109.13
Accurijcyof Type 727 Tape WritinglReading. In recent sales proposals,
IBM has claimed that experience with Type 727 Tape Units in the first
15-20 Type 705 installations has indicated that~ightly less than one
unreadable tape error OCCUrS per 8-hour shift. The Type 729, announced
for use with the Type 709 EDPM, is a modified 727 which includes an immediate reading of newly-written tapes as its major change •
.EQI the 727: This is a commendable performance record, if true.
It
is much better than has been experienced by some known 702/705 installations.
Against the 727: If the Type .727 Tape Units and magnetic tape are this
good, why has IBM found it expedient to revamp the tape writing mechanism to give an immediate re-read for the avowed purpose of proving
legibility? If the 727 is so accurate, is it worth another $100 a
month per tape unit to install additional checking? For sales purposes, note that the 729 cannot be used in central processor operations with the 650, 702, 704 or 705, which require a Type 727. The
fact that the 729 is incorporating this additional checking feature
can be used, however, to combat claims of the accuracy of the Type 7270
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
1109014
Is ttperfect U Tape Possible? The major reason for including the read ... check
feature in the 729-709 is probably not so much to prove that tape recording
is good, but to provide a means of detecting flaws on the tapeo (1108).
In. other words, the principal use of the feature admits that, even if
tape is perfect when it leaves the factory, it acquires flaws in use. Unless these flaws can be passed over in some manner, two shorter pieces of
tape result. With the 729/709, a means is provided for passing over these
bad spots; this is not possible in the other eguipments, but ALL OF THEM
USE EXACTLY THE SAME TAPE.
Against the 727: If IBM's present tape performance is so good (see
1109013 for claims made), why has the company found it necessary to
provide a means of bad spot detection in a system that won't be in
use until late 1958? The fact that they HAVE incorporated the feature
is a definite indication that today's tape ISN'T AS GOOD as they claim
and it does develop: flawso The inclusion of the read-check feature in
the 729/709 and provision for program-controlled passing over of bad·
spots is the best proof available that Type 727 tape performance, and
~ availability of perfect tape, IS NOT SATISFACfORY.
Revised 5/15/57
1109.50
1109.50
SALES POINTS ON TYPE 729 MAGNETIC TAPE UNIT
Pending operating experience with this tape unit, results of uses of the
Type 727 must form the basis for sales evaluationso These sales points
on the 727 are pe~tinent:
1109.01
1109.02
1109.03
1109.51
1109.04
1109.05
1109.06
1109.08
1109.09
1109.10
1109.11
1109012
1109.9 (last page of this section)
Necessity of Readability Check. The Type 729 functionally differs from
the Type 727 only in having the "split head" that reads a newly-written
record immediately after recording it on tape; this costs an additional
$100 monthly per tape unit. Recent IBM sales proposals for Type 705 installations, using the 727 tape unit, stress the point that experience
with the first 15-20 such installations are having somewhat less than one
unreadable tape error per 8-hour shift.
For the 729: This is a nice feature, provided it performs a useful
function.
Against the 729: If the performance of Type 727 tape units is as good
as claimed in Type 705 proposals, is it worth an extra $100 a month
for each unit to prove tape legibility immediately after writing?
Second, how much of IBM's tape troubles are caused by inability to
read a tape on other than the unit on which originally prepared--and
this is a known occurrence which happens quite frequentlyo Reading
difficulties caused by tape unit "incompatibilities" will not be corrected simply by proving that a tape is good; the fact that a number
of "illegible" records can be read properly merely by switching the
reel to another tape unit is proof of that.
Original 2/15/57
1109.9
(1)
1109.9
Unchecked Control Errors on the 7270 In all EDPMs the control of
tape units--reading, writing, rewinding, backspacing, etc.-- is through
use of normal machine instructions. All IBM equipments using the 727
are, essentially, completely unchecked insofar as verifying the accuracy
of instruction execution is concerned. Univac verifies that the correct
operation is actually performed--another feature of Univacvs justlyfamous t'self-checking t• circuitry.
For the 727. Because checking is basically a function of main frame
circuitry, there are no evident 727 advantages in omitting it.
Against the 727. Each of the types of "busts" listed are known to
have occurred in 702 installations; because the other types of
equipments use the same basic unchecked method of operating tape
units, comparable performance can be expected.
1.
2.
3.
4.
5~
Failure to recognize the reflective spot at the end of the tape
while writing. This is an unchecked photoelectric sensing which
gives only one signal when working properly. It doesnVt always
give the signal. See also item 2 in 1109.09.
Failure to recognize the tape mark when reading. This likewise
is a single unchecked signal. However, it is rather easily
corrected--at the expense of some programming--by writing two
or three tape marks after the last information record. The
equipment is bound to pick up one of them.
Failure to execute a "backspace" instructiono Although the 727
can never flread backwards," it can be backspaced one record at
a timea The instruction is not always executed, with the result
that it is perfectly possible to completely drop a record from
a master file during an updating operation. It is understood
that many customers are allergic to thiso It is certain to
happen--particularly over a period of time--unless elaborate
and time-consuming programming is incorporated to make sure
that a backspace instruction actually backspaces. One method
is block or record numbering--which takes both program steps and
computer time.
Writing one output record two or more times. This has happened
and is possible in two ways. First is failure of the instruction address counter to advance to the next instruction address-unlike Univac, this is unchecked in at least the 702 or 705.
The result is that an instruction is executed twice. If it's a
t'write" instruction, an output block is duplicated. The seriousness of the error depends on the file involved; it may be only
a source of annoyance in a future machine run; it could also
result in posting $5,000 to my checking account twice" The
second cause of duplicate output records results when a 'twri te ft
error is indicated and the tape fails to backspace before it is
rewritten. This usually is more of an annoyance than anything
else; of course, the odds are good that the tape is to print a
management report which looks rather loused up in finished formo
This is particularly embarrassing when the output block is repeated four times--which has happened.
Failure to select the proper tape unito Because the equipments
do not verify that the tape unit specified by the instruction is
the one actually connected, it is possible to write 0.0 an erroneous tape (effectively losing information which belongs on another output), or reading in from an incorrect tape. Use of
1109.9
(Continued)
the '''protective rings" will eliminate the possibility of
accidentally writing on a protected reel, but the other
chances of error remain. This type of error often merely
"stallstt: the machine, because the erroneous address may
apply to a non-existent tape unit or tries to Itwrite" on
a protected unit which can only be in a "read" status.
1120
IBM TYPE 652 TAPE AND RAMAC CONTROL UNIT
The Type 652 is a control unit used for three separate purposes, alone
or in combination, in the IBM 650 EDPM system.
a.
b.
c.
To connect the Type 727 Tape Unit;
To connect the Type 355 Random Access Memory (RAMAC);
To connect the Type 838 Inquiry Keyboard.
Use of either magnetic tapes or RAMACs in the 650 system requires also
the inclusion of a Type 653 Record Storage Unit (see 1125) and, to be
technically correct, the 653 connects the tape units or RAMACs to the
high-speed storage element contained in the 653. Type 838 Inquiry Keyboards can be used only if the 650 installation also includes RAMACs.
1120.1
Monthly Rentale Rental varies depending upon the type of configuration
involved.
652 for magnetic tape units only.
s
•
652 for RAMACs only •
•
•
e
e
_
•
•
•
•
:
652 For magnetic tape units and RAMACs • • • • • • • •
Each Inquiry Control Unit (maximum two), additional.
$ 1,050
975
1,350
300
Monthly rental for a 652 adapted only for RAMACs will be $975, $1,275,
or $1,575, for none, one or two Inqui;ry Control Units; if also equipped
to handle magnetic tapes, the corresponding rentals are $1,350, $1,650,
or $1,950.
1121
CHARACTERISTICS OF THE TYPE 652
The basic purpose of the 652 is to provide the power and control signals
necessary to handle magnetic tape .units, RAMACs and Inquiry Keyboards.
It does not need to be considered in the writing of programs using magnetic tape units or RAMACs, nor in the operation of Inquiry Keyboards;
however, it is involved in the initial setup of the latter. It can
control a maximum of six Type 727 Tape Units, four Type 355 RAMACs and
ten Type 838 Inquiry Keyboards through one or the Inquiry Control Units.
Any combination, up to the maximum of each type of unit indicated, is
permissible, provided that the 652 is equipped to handle them (see 1120.1).
1121.1
Characteristics of the Type 652 -- Magnetic Tapes
Number of Type 652 which can be connected
to one Type 650 (through the 653)
Number of Type 727s connected to one 652
Tape Unit Addresses
Method of Address Selection
Simultaneous read/write operations
Restrictions on number of tapes which can
read or write
Simultaneous writing
1
1 to 6 (maximum)
8010 through 8015.
Rotary switch on 727$ for
units position; "801" is
built in
.
No; read only or write only
None, within limit of 6
Two tapes, by setting both
to same address
1121.2
1121.2 Characteristics of the Type 652
1121.3
RAMACs.
Refer to 2120.
Characteristics of the Type 652
Inquiry Keyboards.
This also discusses the Inquiry Control Units.
Refer to 7110.
For a fuller discussion, refer to sections on magnetic tapes, RAMACs,
Inquiry Keyboards, the 650 and the 650 RAMAC.
1125
1125
IBM TYPE 653 RECORD STORAGE UNIT
The IBM Type 653 Record (or High-Speed) Storage Unit must be used if
magnetic tapes or RAMACs are attached to a Type 650 EDPM, but may also
be used, without either of these, as an extension of the storage
capacity of the Type 650 itself. If used with tapes or RAMACs, a Type
652 Control Unit (see 1120) is also required. Only one Type 653 can be
associated with one Type 650.
1125.1 Monthly rental:
1126
CHARACTERISTICS OF THE RECORD (HIGH SPEED) STORAGE UNIT
The characteristics listed below are common to all uses of the Type 653.
More detailed discussion of its use with Type 727 tape units is contained in 1127; with Type 355 RAMACs in 2120 and 6150; with the 650
processor alone in 6100.
Type of storage:
Character Capacity:
Addressing Facilities:
Address Range:
Input to High-Speed Storage:
Output from High-Speed Storage:
Multiplexing Ability:
Magnetic core
60 words (600 decimal digit plus 60
signs)
Each word individually addressable
9000 through 9059
(1) Directly from magnetic tape or RAMAC
(2) From any other input device through
the input buffer
(3) One word transfers from distributor
and accumulator
(4) Block transfers from general storage
on the 650 drum
(1) Directly to magnetic tape or RAMAC
(2) To any other output device through
the output buffer
(3) One word transfers to distributor,
accumulator or program register
(4) Block transfers to general storage
None. The 653 can carryon only one
operation at a time. I.e., if reading
from one tape, it cannot write on
another, cannot be transferring data
to or from a RAMAC, nor do any other
type of operationo
1126.1
1126.1
Block Diagram of Type 652-653 Association in Type 650 EDPM
This diagram shows the relationships of the 653, the 652 Control Unit,
tape units and RAMACs, with information flow paths, in the 650 system.
Tape
Control
Unit
652
Recorq
Storage
Unit
653
Input
Buffers
Ou}put
Buf ers
Accumulator
Distributor
______________~ Program
Register
TYPE 650
General
Storage
1127
1127
USE OF TYPE 653 IN MAGNETIC TAPE OPERATIONS WITH TYPE 650 EDPM
-When used with Type 727 tape units in 650 operations, the Type 653 HighSpeed Storage Unit serves both as tape buffer and as an extension of the
general storage (the drum) of the machine. It cannot serve both functions simultaneously; when tape is being written or read, the 653 cannot
be used in carrying on other instructions, which must be confined to
the general smrage on the drum. The length of time that High-Speed
Storage is tied up depends upon the particular type of operation involved in tape reading or writing; typical times are given in 1127.2.
1127.1
Length of Tape Records. Refer to 1103.3 for description of tape formats
used. The point from which a tape record is written onto tape, or into
which it is read, is determined by a timing ring which serves essentially, as a starting point in the 60-word magnetic code high-speed
storage. This timing ring can be set by several means (not discussed
here) to any address between 9000 and 9059; in both input and output
operations, the data transfers begins with the address of the timing
ring and continues through address 9059~ Thus numeric information can
be read or written in full-word increments of any length from 1 to 60;
alphanumeric information is subject to the restriction that it must be
in increments which are multiples of ten words.
1127.2
Operating Times for Tape Input-Output Functionso The use of magnetic
tapes with the 650 imposes the necessity of certain control interlock
functions during the execution of the various types of tape instructions.
The following table indicates the times necessary for carrying out the
types of operations indicated, and also shows the period of time during
which a tape unit, the 652 tape control unit,and the 653 storage unit
are not available for other work.. The times given do not include that
necessary for the instruction initiating the action; this averages about
0.3 ms. All times are in milliseconds.
Read/Write Tape:
10
I
0.6SN, where N is number of characters in the
tape record
A 60-word numeric record
requires 50.S ms to read or write.
All tape units are unavailable during this period.
The 652 is unavailable during this period.
The 653 is unavailable during this period.
11.0 ms. The 652 and all tape units are unavailable
during this period. The 653 is available throughout
the period.
If tape in write status: 46 I 0.06SN ms;
It
tf
in read status:
30 I 0 .. 06SN ms;
All tape units except the one addressed and the
652 and 653 are available during this period.
Maximum 1.2 minutesQ All tape units except the one
addressed and the 652 and 653 are available during
this period ..
G
Write Tape Mark:
Backspace Tape:
Rewind Tape:
In these remarks, the term "unavailable" means that an instruction addressed to any unit during the time period indicated will cause the 650
tot'hang up" on that instruction until the current operation is completed.
The maximum tape transfer rate of the 650 is about 11,SOO characters per
second, combined read and write, or about 5,900 each. Effective transfer
rate is much less, because computation may not overlap and, at any event,
transfers between general and high-speed storage may be required ..
1140
1140
The Type 754 is an unbuffered Tape Control Unit providing power and
control functions for up to ten Type 727 Tape Units in 702/705 EDPM
installations. In either type of system, up to ten 754s, or combinations
of 754s, 760s, and 777s, may be connected; each of the 754s in use
may have from one to ten tape units. The maximum tape unit complement of
the 702/705 series is 100, achieved with ten 754s.
1140.1
Monthly rental:
1141
BASIC CHARACTERISTICS AND USE OF THE Type 754
$2,000.
See 1103.4 and 1103.6 for explanation of tape formats with the 702 and
705. The 754 provides unbuffered communication between the 702/705 memory and tape units. Because it does not include a buffer, use of the
754 means that input and output do not overlap computation. In addition,
the 702 is not able to read and write tape simultaneously; consequently,
reading, writing and computation proceed in sequence, with none of the
three overlapping. The 705 is able to read on one tape and write on
another simultaneously, but nothing else can occur until the longer of
the two operations has been completed. With the 754, computation never
overlaps input-output operations.
The principal advantage of an unbuffered tape unit is that it permits
extremely long tape records; for example, the first sorting routine developed by IBM generated strings of somewhqt mOre than 8,000 characters
which were written as one tape record. This not only increases the total
character capacity of a tape unit, but also reduces input/output time by
cutting down the number of inter-record gaps which must be traversed in
a tape operation.
One characteristic of the 754 is worth mentioning: In use, it is not
necessary for the tape units to be "stopped" in order for a read/write
instruction to be executed; regardless of where it is in the stop cycle, the
next instruction commences immediately. Thus, as soon as the last character of a tape record has been read into memory, tests can be initiated
to determine whether it is required for processing; the 702/705 are not
held up until the tape has stopped. Such a test may take less than a
millisecond, at the end of which time a combined read/write instruction
(in the 705) will cause both tapes to immediately resume full speed, even
though not stopped. The stop curve is such that it always takes 10 ms
to cover any remaining portion of the inter-record gap. Therefore,instead of starting and stopping the tape in the gap, which would require
a minimum of about 16.4 ms, it is often possible to cover it in about
11 ms, one used for computing (during which time the tape is starting to
stop) and 10 for covering the remaining portion of the gap.
The time required to read, write, or read/write is 10 I 0.067N ms, where
N is the number of characters involved (in read/write, the number of
characters in the longer of the two records). Nothing is done in
this period; the time is 77 ms for a 1,000-character record.
1145
1145
IBM TYPE 755 TAPE CONTROL UNIT
The Type 755 is a Tape Control Unit connecting from one to eight Type
729 Tape Units to one channel of a Type 766 Data Synchronizer in 709 EDPM
operations. Its basic function is to provide power and a few control and
switching circuitry functions to the tape units in on-line processing.
It does not need to be taken into account in programming and has no operating features requiring special comment.
A separate 755 is required for each Data Synchronizer channel having
tapes attached, regardless of the number associated with each channel.
Purchase:
Rental:
$108,000
1,800
These prices apply to the Modell, which is the only one so far announcedo
1145.2 Discussion of Useo The unit is of no programming or operating importance.
Refer to section 1160, Type 766 Data Synchronizer, for a discussion and
evaluation of the use of magnetic tapes in a 709 system.
No sales points are included for this equipmento
Original 2/15/57
1149
SALES POINTS ON THE TYPE 754 TAPE CONTROL-lltlII
IBM does not appear to be including 754s in many of its current proposals.
this is· due to the fact that it is unbuffered, and also to the fact that
the cost of the 777 Tape Record Coordinator is only $1,000 a month more.
Occasionally, a 754 will be included in sales proposals where total rental
~s a critical competitive factor; it permits proposing a system with ten
tape units (rather then the eight maximum with one 777) at a cost less than
a comparable Univac II. In some instances, has proposed as few as six or
seven tape units with one 754 to get the price down to a figure close to a
Univac I. The following points should be brought out when a 754 control
unit is proposed with a 705; in general, these points are true even though
one or more 777s are also proposed.
1149.01
Inability to Read/Write and Compute Simultaneously. With the 754, the
705 can write from one area of memory and read into another area simultaneously, but can do nothing else until both operations are completed.
All computing must be done outside of tape time. This includes, as a
normal minimum, all tape read, write and error test instructions, testing
of data to determine if the record is to be processed, possible block
reassembly into a new output, and either internal memory transfers to
move the input to an output area, or modification of input/output instructions to permit writing out of the area into which reading occurred.
The amount of this computing time varies greatly with applications.
Against the 754. Simultaneous read/write operations are no novelty-Univac has had it since the first one was built, and of course also
computes simultaneously. The necessity of computing outside tape
time adds--sometimes greatly-- to total processing time.
1149.02 Comparative Processing Speeds of a 705-754 and. Univac II. The effective
speed of a 705-754 is dependent not only upon the amount of computing
inv~lved, but also upon the length of the grouped tape records.
Greater
tape speed is attained by grouping several records into one long tape
record, cutting down the number of gaps which must be passed over; of
course, the longer the record, the more memory must be aside to hold it
and the greater the amount of computing pet tape record. A nQrmal maximum is about 1,000 characters, partly because some customers expect to
get 777s and set up records initially on that basi.s.. The following table
shows typical speeds possible, in numbers of characters per second, for
Univac II and the 705-754. Various length tape records are included for
the 705 (UNIVAC always uses a 72o-character block), and times are shown
for instructions varying from 50 to 500 per tape record.. The amount of
computing time assumes an average of 0.14 ms per instruction for Univac
II and 0.07 for the 705; this is normally not valid but the 705 does
execute individual instructions at an average rate about twice as fast
as Univac II. This does not mean, of course, that it will perform a
required processing operation twice as fast, internally, as Univac II.
1149.02
1149.02
(Continued)
Average Processing SQeeds of Univac II and the IBM 705
Using the T~Qe 754 TaQe Control Unit
(Characters per Second)
Number of
InstrVns
50
100
200
300
400
500
Univac II
14,300
14,300
14,300
14,300
11,600
9,500
Length of TaQe Records! 705-754
1000
600
800
400
200
7,450 9,900 11,100 12,100 12,400
6,600 9,100 10,500 11,500 11,900
5,350 7,850 9,300 10,400 11,000
4,500 6,900 8,400 9,600 10,200
9,500
3,900 6,150 7,700 8,800
8,900
3,400 5,500 7,050 8,200
Estimating the number of instructions required to process one block is
frequently difficult, particularly in sales proposals. As a reference
point, it may be noted that the typical Univac II block requires about
150 instructions for necessary housekeeping, block reassembly and tests
to determine if any item in the block must be processedo The 705 will
take fully as many. Actual processing is in addition to this, and
varies considerably; it is always additional time in the 705-754, while
Univac II can execute a total of about 325 (or something like 175 aside
from housekeeping) within tape time. In connection with the 705 speeds
noted above, see also the next sectiono
About the 7540 These two points should be brought out:
(1) Computing is always in addition to tape time, and tape read/write
speeds are slower than Univac II.
(2) Univac II, on the average will out-perform a 705-754 combination
by at least 35-4~fo, and it may run more.
1149.03
Effect of Variations in Length of Tape Records. The above speeds are
based upon 705 tape records of identical lengths, a condition automatic
in Univac but not in the 705~ Because the 705 is tied up in the tapememory transfers until the longer of the two unequal-length records has
been handled, it is obviously desi.rable to have them all the same length.
Many business files, of course, have records with variable amounts of
detail information appended to them, and IBM has laid great stress on
the ability of the 705 to handle variable-length words and records. If
records are variable in length, maintenance of a fixed tape record must
be programmed; usually reassembly of an input tape record into parts of
the outputs must be provided for, because most file maintenance operations have to provide for insertion and deletion of information. IBM
will doubtless point out that its high-speed internal memory transfer
makes this easy and fast to accomplish. They will not point out the
details necessary to use the high-speed transfer; these seriouslY slow
down the operation. Refer to 5140 for a fuller discussion of the ramifications of the high-speed internal memory transfer~
If tape records are allowed to vary in length, the total time required
to read/write a file is greater than a time calculated on the basis of
the average length record; how much greater depends upon the variations
1149.03
1149.03
(Continued)
in length and their distribution. For example, the time needed to
process a file consisting of alternating 1000 and 100 character tape
records is exactly the same as for a file of all 1000-character records.
Typically, the time increase, as compared with average record lengths,
will range from 15-25%. The alternative is reassembly internally in fixedlength tape records, which requires computing time. 'There is evidently
a balance point; on one side of it the increased tape time is less than
the computing necessary for reassembly, and on the other side the converse
is true. Determining where this point is for any given file is extremely
difficult, if not impossible.
Against the 754: Point out the above facts; it is suggested that the
indeterminateness of the increased processing time be emphasized.
In practice, it is almost impossible to estimate the time for a file
maintenance run with any degree of reliability, even if itis completely programmed. These restrictions and considerations do not
exist in Univac.
1149.Q1 Number of Tape Units Which Can Be Controlledo With ten 754s, the 705
has a maximum of 100 tape units which can be connected. Univac II is
limited to 16. The fact that the 705 has a capacity for more may be a
factor in some sales proposals.
For the 754: The 705 system can handle more tape units than the Univac
systems.
Against the 754: Most Univac II orders do not include even the 16
maximum available. The number which can be used profitably is in
large measure a function of the total memory capacity available;
if more inputs and outputs are contemplated (and why have more tape
units if this is not true?), it is self-evident that more memory is
needed for instructions to handle them and, in many instances, more
memory will be needed for data storage. The optimum number of tape
units for any given installations must be predicated upon consideration of all applications involved; Available knowledge indicates
that, for EDPMs of the scale of the 705 and Univac II, from 10 to
16 are all that can be economically justified~
1150
1150
IBM TYPE 760 CONTROL AND STORAGE UNIT
The Type 760 Control and Storage Unit is primarily designed as a control
and buffer storage unit for off-line printing operations using anyone of
the Types 719, 720 or 730 high-speed printers. Additionally, it can be
used in 702/705 central processor operations as a tape/printer control
and buffer storage unit. These configurations are possible:
1.
Off-Line Operations. One 760 with one HSP (719, 720 or 730) and
either one or two Type 727 Magnetic Tape Unitso
2.
On-Line
a. One
b. One
c. One
Operations With 702 or 705.
760 with one or two Types 727 Tape Units.
760 with one HSP (any type).
760 with one HSP and either one or two Type 727 Tape Units.
The use of the 760 is mandatory with the IBM 500-and 1000-line a minute
printers (Types 719, 720 and 730), either on-line or off-linea Its use
for this purpose is discussed in section 4130. This section is limited
largely to its use as a tape control unit in on-line operations. In
either method of operation, all normal characteristics of the Type 727
Tape Units (1100) remain unchanged.
1150.1
Monthly Rental:
1151
CHARACTERISTICS OF THE TYPE 760
$1,800.
Number of Type 727 Tape Units controlled
Number of printers controlled
Type of storage (buffer)
Storage capacity
Character transfer rate:
Memory and 760 storage
Tape and 760 storage
Checking of information transfers
EDPMs used with in on-line operations
Number of 760s possible in on-line
operations
Permanency of stored information
1 or 2 (maximum)
1 only
Magnetic core
1,000 702/705 characters
0~023 ms/character
0.067 ms/character
Yes
702 or 705
10; or total of ten in any
combination with Type 754
Tape Control Units
Non-destructive read-out;
character by character replacement on read-in
1152
1152
ON-LINE USE OF THE TYPE 760 WITH 702 or 705 EDPMs
The Type 760 Control and Storage Unit may be used in 702 or 705 CPU
operations as a buffer storage unit controlling one or two Type 727 Tape
Units; it can serve this function with or without an HSP also attached.
This section describes the method of operation as a tape control unit
and also introduces the instructions actuating an HSP in on-line operation.
1152.1
Essential Operating Features of the Type 760. This sub-section outlines
the basic features of the 760 which pertain to on-line use. Unless
otherwise noted, 702 and 705 uses are identical.
1152.11 Method of Addressing the 760, Tape Units and Printers. The address
selecting a tape unit or printer controlled by a specific 760 is of the
form "02XY, tt as the address of a ttSELECTt~ instruction.
tt02 tt designates a tape control unit;
ttX" is a digit, from 0 to 9, designating the specific 760 involved.
It differs from that assigned to any other 760 or to a Type 754
or (probably) Type 774. Thus a maximum of ten such control
units, in any combination, is possible;
tty.' designates the specific tape unit or printer:
2 or 7 - Tape unit
4
- Printer
Invalid; anything other than 2, 4 or 7 in units posiOthers
tion of address causes a machine stop.
1152.12 Storage Counter. The 760 contains a counter device that operates from
000 to 999 for the 1,000 character positions of buffer storage. It is
used to permit loading the storage unit with several successive records
from memory; the method is discussed in the sections outlin:ing'.instructions
used with the 760.
1152.13 Control Mark. The end of information in the storage unit is indicated
by a special character code called the control mark. Its function is
similar to that of the record mark in the 702 (1103.4) or the group mark
in the 705 (1103.6). Available information does not indicate the bit.
combination forming the control mark, but apparently it is a new one not
used in any other component of the 702/705 systems. Its use is covered
in the following sections on 760 instructions and summarized following
that discussion.
1152.14 Checking of Data Transfers. All data transfers into and out of the 760
storage unit are given redundancy bit tests and tape records read in also
receive the check character test. Unlike the Type 777 IRC, the check
character is not placed into the buffer storage. Error indications are:
1.
Memory to 760:
2.
3.
760 to Memory:
Tape to 760
4.
760 to Tape
5.
760 to HSP
0902 Read-Write Check Indicator. Usually an error
also turns on 0901 Machine Check Indicator.
0902 Read-Write Check Indicator.
0903 Record Check Indicator. This transfer also
verifies the check character on the tape record.
0903 Record Check Indicator. The check character is
also created, but of course not checked.
0903 Record Check Indicator.
1152.14
1152.14
(Continued)
In all cases, either automatic machine stop or program-controlled testing
and correction is available. Error indications for the first two types
of transfers listed are available immediately upon completion of the
transfer; for the last three, the check indicator turns on during execution of the next read or write instruction, which may involve a completely
different control unit.
As a matter of interest in the logic of the 702/705 design, no·te that a
tape-to-760 error is indicated by an 0903 Record Check Indicaior; a tapeto-777 error, by an 0902 Read-Write Check Indicator. Literature available does not explain the reason for the two different approaches to
handling an identical type of machine mistake
o
1152.15
Tape Unit n'Early Start tt, Feature.. Like the Type. 777 IRC, it is poss.ible
to start a tape in motion during the time a memory-to-760 data transfer
is being made, thereby permitting a partial overlap of operations. The
feature can be used when the transfer from buffer to memory can be completed in 4 ms or less (early start on "read" tape), or memory to buffer
in 8 ms or less (ear 1y s tart on "write ft. tape). This is time enough to
handle 173 and 347 characters, respectively. (IBM uses 170 and 340 as
approximations). The reason for the different time intervals is the same
as that explained in 1182.21 and 1182.22.
1152.5
Instructions Used in
e 760 0 erations. Follr new instructions are
introduced to the 702 705 command list when a 760 is installed. The
illustrative examples used in the following sections have addreses 0212
and 0217 for tapes and 0214 for the HSP; thus they refer to a 760 with
address "1ft. Error-testing instructions are omitted, but must of course
be included in working programs.
1152.21
760-to-Memory Transfers. This type transfer is normally associated with
an input tape which is to be read into the 760 buffer after its contents
have been placed in memory.. Assume 0212 is this input. The 760-tomemory transfer then is accomplished by this set of instructions:
SELECT
READ
0212
m
Selects 760 control unit "Itt and tape unit tt2ft
Transfers one record from 760 buffer storage into
702/705 memory beginning at address m.
The transfer is terminated by a record mark, or the control mark following the last record, in the 760 buffer. _B~cause several records may be
contained in the 760, several "read" instructions are necessary to
transfer all of them. Each such "read" transfers one record into the
memory location designated; rather evidently, successive records in the
buffer do not have to be read into successive memory locations. In the
702, the record mark is transferred into memory following the last information character; the control mark following the last record in the
buffer is changed automatically into a record mark. In the 705, it is
not transferred; the record or control mark terminates the operation,
but the last character moved is the one immediately preceding it in the
760 buffer.
1152.21
1152.21
(Continued)
760 Buffer Storage
Record#3 Record #4
Ml
M2
M3
Ml
{.--..IR_ec._#1______
I R_ec._#2_____
I Re_c.#_3~I~
702 Memory Transfer
\ 1Rec.
#4
Rec. #5
Rec. #6
705 Memory Transfer
The nature of the successive transfers in both the 702 and 705 is shown
in the above illustration. This is illustrative of the transfer--rather
obviously one 760 cannot be attached simultaneously to both a 702 and 705.
Considering the 702 example, the first "read tt instruction moves the first
record in the 760 into memory beginning at address MI. At its completion,
the 760 storage counter (1152.12) is set at the position of the first
character of Record #2, and, when the next "read" instruction is executed,
the 760 begins transferring at that character position and continues Until the next record mark is reached, and so on:
SELECT
Read first record into Ml
Read second record into M2' etc.
READ
READ
Particularly in 705 usage, several successive "read" instructions are
normally necessary to transfer the full buffer into memory; this can also
be true in the 702, as will become apparent shortly.
1152.22 Memory-to-760 Transfers. This is usually associated with one of the tapes
or the printer as ultimate destination of the information; assume it is
tape unit 0217. The memory-to-760 transfer requires:
SELECT
WRITE
0217
m
Select 760 control unit ttltt and tape unit "7"
Transfer into 760 buffer beginning at memory
location m
In the write-out, the group mark in the 705 terminates the operation,
with several records separated by record marks usuailyinvolved. In the
702, the record mark terminates the transfer. In a manner similar to
reading, it is possible to load several successive records into the 760
by a series of ttwrite tl instructions, a procedure which would be applicable
particularly to the 702. The storage counter controls the assembly of
each record into the successively higher positions of the 760 buffer.
1152.23
CONTROL 0006 (RWS 0006). This new control instruction is used to read
in a record from tape, or to write a record on tape or printer. It normally is preceded by a previous "read" or "write" instruction which determines the meaning of "RWS 0006 (Eead/~ri te §.torage). Instruction sequences
are:
I~
1152.23
1152.23
(Continued)
Reading
SELECT
READ
RWS
0212
m
0006
Select input tape
Read (last) record from 760 buffer into memory
Start tape unit to read next record or grouped
record into 760 buffer
0217
m
0006
Select output tape
Write record from memory into 760 buffer
Write contents of 760 buffer onto tape
Writing
SELECT
WRITE
RWS
If the printer (0214) is selected in a "write tt operation, one or mOre
lines are printed from the 760 buffer. These points are apropos:
1.
RWS 0006 resets the 760 storage counter to 000 at the beginning and
completion of the data transfer. Therefore it always reads into or
writes from the beginning of the buffer and leaves the counter reset
at 000 when the read or write is completed.
2.
RWS 0006 causes an automatic conversion of the last record mark in
the buffer to a control mark, which terminates a writing operation.
In the 705, the group mark terminating a memory-to-760 transfer is
changed to a control mark by this instruction (individual record
marks within grouped records are unaltered).
3.
IBM notes that the 760 can be used in a "tape copyingtt routine without putting the .information through memory:
SELECT
WRITE
SELECT
RWS
SELECT
RWS
0217
X
0212
0006
0217
0006
Select output tape
Any address to put tape in "write" status
Select input tape
Load 760 control unit from input tape
Select output tape
Write 760 buffer contents on output tape
The point of interest here is that the reading from "0212" is the
only known instance in which IBM refers to a tape unitOs reading or
writing without being specifically instructed as to which is involved.
1152.24 CONTROL 0007 (RWT 0007). This instruction is used only with tape units
and initiates the early start feature similar to that in the Type 777
TRG. In use, it allows the next "read" instruction to transfer one record
from the 760 into memory while the tape is accelerating and before the
input record replaces the contents of the 760 buffer. In "writing, ,t one
more record can be loaded into the 760 before the actual 760-to-tape
transfer starts, but while the tape is accelerating to full speed. In
either case, a typical instruction sequence is:
SELECT
0212
RWT
0007
READ (WRITE) m
Select input (output) tape
Start selected tape in motion
Read (write) one record from (to) 760 buffer,
memory address m
1152.24
1152.24
(Continued)
The reading (or writing) begins in the buffer character position designated by the storage address counter, which is not reset to 000 until
after completing the read or write instruction following "RWT 0007."
(RWT -- Read/~rite Iape). Time permits about 170 characters to be read
into memory (4 ms maximum at 0.023 ms/character), or 340 to be written
from memory (8 ms maximum). The time saved is the difference between
standard tape start time of 10 ms and the number of milliseconds required
for the memory-760 transfer.
If more than one read or write instruction follows an RWT 0007, the
second is "held up" until the tape-760 transfer is completed; because
the storage counter is reset to 000 after this transfer, the following
read/write begins at this buffer position.
Parenthetically, there is no information as to what happens if the "read"
or "write" takes longer than the 4 or 8 ms maximum. Because several possibilities exist, speculation is rather pointless~
1152.25
CONTROL 0008 (RST 0008). This instruction resets (RST -- Re§eI) the
storage counter of the selected 760 to 000 without altering the buffer
contents. This, coupled with the necessary number of t'dummy" reads, permits any desired record in the buffer to be read into memory repeatedly
without using a tape unit, or to substitute a revised record into the
buffer before writing a new output tape. This may save some time in 760memory transfers, although the entire contents of the buffer through the
~ecord involved must be handled at 0.023 ms per character.
This instruction also completes the methods of resetting the storage
counter to 000:
1.
2.
3.
By an RST 0008 instruction;
At the beginning and completion of an RWS 0006 instruction (automatic);
After the read or write instruction following an RWT 0007 (automatic).
It will be noted that it is not possible to read several tape records
and accumulate them in the buffer; each begins at 000 and destroys the
previous contents. Likewise, several records in the buffer cannot be
written out as individual tape records.
1152.26
CONTROL
with an
to both
(PTW --
0009 (PTW 0009). The last of the new instructions is used only
on-line high-speed printer (719, 720 or 730) when it is desired
print and record on tape the information in the 760 buffer.
~inter and Iape ~rite). The sequence of instructions is:
SELECT
WRITE
RWS
SELECT
PTW
0214
m
0006
0217
0009
Select the printer
Move output record (or last one) into 760
Print record group
Select output tape
Write record group on output tape
In a 702 program, several "wri te tt instl'uctions may precede the RWS 0006.
Although instruction manuals are not specific, it is probable that the
1152.26
1152.26
(Continued)
printing and tape writing are sequential in execution, rather than occurring simultaneously. The reason for the "PTW 0009," rather than an
"RWS 0006, tf. instruction for the tape record writing arises from the fact
that, in printer operations, the 760 buffer is not loaded solidly, but
in sectiQns. (See 4130 for details). Use of "PTW 0009" permits skipping
the gaps between sections and recording on tape only the actual data
involved in printing.
1152.3
Surrunary of Type 760 Instructions and Operating Times. The following
table surrunarizes the times required to execute instructions used with
the 760; tape and printer times are als 0 shown.
Operating Time in Milliseconds
Instr.
READ
Remarks
760 to memory
702
.13S!.023N l
705
.034!.023N 1
WRITE
Memory to 760
.13S!.023N l
.034!.023N 1
RWS 0006
RWT 0007
760 and Tape/Prntr
760 and Tape
0.163
0.051
0.163
0.051
RST OOOS
PTW 0009
Reset storage counter
Print & write tape
0.163
0.163
0.051
0.051
Nl
N2
*
***--
**
1152.4
I
Tape
Printer
*
*
*
*
10!.067N2 60/1ine**
10!.067N2*** ------
10!.067N2
---
60/1ine**
Number of characters in a single record
Number of characters in a grouped record
Not available during 702/705 operating time
Maximum of ten print lines for one instruction
The following uread u instruction, not to exceed 4 ms, or tlwri te, ft,
not to exceed S ms, occurs during tape time
Surrunary of Handling of Record, Grouped and Control Marks. The use of three
separate character codes to signify ends of records, or groups of them,
and various ways in which they are handled, can cause considerable confusion. This section summarizes their status and use in 760 operations.
1.
Units With Which Used:
a. Control Mark: Peculiar to the 760 storage unit in both 702 and
705 use, on-line and off-line. It never appears in memory_
b. Group Mark: Peculiar to the 705 and appears in the 76~ buffer.
c. Record Mark: Used in 702, 705 and 760.
2.
Tape-to 760 Transfers. The inter-record gap on tape terminates the
tape read. A control mark is automatically inserted in the 760 buffer immediately following the last character of the tape record.
Record marks on tape are handled as any other valid character. There
is no difference in tape reading for either the 702 or 705.
3.
760-to-Tape Transfers. A RWS 0006 or RWT 0007 instruction addressed
to a tape in a write status automatically causes a change in the last
record mark in the 760 buffer (702), or the group mark (705), to a
control mark. The control mark terminates the writing operation and
1152.4
1152.4
(Continued)
causes the inter-record gap to be created, but is not itself placed
on tape. Record marks in the buffer records are handled as any other
valid character code.
4.
760 Buffer-to-Memory Transfers. In both 702 and 705 use, a record
mark following each record in the buffer, or the control mark following the last one, terminates the transfer.
a.
b.
5.
In the 702" a record mark is automatically inserted in memory
following the last data character. The control mark is automatically converted to a record mark in memory.
In the 705, the record mark is not inserted in memory. The last
character transferred is that immediately preceding the record or
control mark in the buffer.
Memory-to-760-Buffer Transfers.
a.
b.
In the 702, a write operation is terminated by a record mark,
which is transferred into the 760 buffer. Several records can
be loaded successively.
In the 705, a write operation is terminated by a group mark,
which is transferred into the 760 buffer. Record marks within
several grouped records are handled as any other valid character.
The group mark of the 705, or the last record mark of the 702, which
appear in the 760 buffer, are automatically converted to control
marks before tape writing or printing commences.
6.
1152.5
760-Buffer-to-Printer Transfers. The control mark is inserted in
the buffer by the RWS 0006 instruction in the same manner as for
tapes. The record mark, .or control mark following the last record,
stops the transfer of characters for each printed line. Neither
type of ttmark" is printed.
Comments on Record, Group and Control Marks With the 760. From the previous sub-section, it is rather obvious that there is considerable t~hard
ware" detail in handling three kinds of record-terminating marks in different manners. It would be interesting to find out from IBM just why
a "control mark"--a completely new character code--is necessary in the
760 when the same function is handled in the Type 777 IRC with the already-existing group mark. (This might be a pertinent question to ask
of prospects who are convinced of the logical beauty of the IBM 705
system. A few more are raised a little later).
1152.51 Lack of Record Termination in the 705. It will be noted that the 760buffer-to-705-memory transfers never include a record mark. This means
they must be program-inserted into input areas, as must the group mark
necessary to terminate subsequent output operations. An inevitable result is that any file read into a 705 via a 760 must consist of fixedlength records, in a constant number per tape grouped record. Overall,
even though the programming is messy, it probably isn't too important,
because the 760 is hardly ideal as a tape control unit in on-line 705
operations. (This use is discussed in more detail in 1153).'
1152.52
(1 )
1152.52
Automatic Conversion to Control Mark. The automatic conversion of a
final record mark or group mark has all the earmarks of a potential
booby-trap. Let's analyze a few possible cases (which, parenthetically,
are all suggested by IBM in its manuals of information).
1.
705 u·sage. The second of a group of five records in the buffer must
be revised and, to save time in memory-to-760 transfers, the storage
counter of the 760 is reset to 000 and a "dummy" read, using a glop
area of memory, sets the counter to the beginning of the second
record. A group mark is program-inserted at the end of the revised
record in memory, and a "write" instruction transfers it into its
proper place in the buffer, which now looks like this:
C
Rec. #5
Rec. #3
Rec. #4
Rec. #1
Rec. #2
M
t
f-
:f:
f
Just for fun, further assume that Record #5 must be worked on: when
re-loaded into the buffer the control mark (elM) is replaced by a
group mark. Now letVs write the grouped records on tape. RWS 0006
changes the group mark to a control mark and the record is written,
stopping when the control mark is reached. Unfortunately, the tape
block now has a group mark terminating Record #2, rather than a record
mark. Sometime in the future this tape will be read in again$ Now
Records #2 and #3 will be transferred into memory together (remember
the group mark doesn't stop a 760-to-memory transfer). But suppose
Record #3 now has some work to be done on itQ Well, obviously it
follows Record #2 on the tape--because we laid out the tape we know
that much. Easy to handleo Reset counter to 000. Dummy read to a
glop area--thereVs Record #1 out of the way. Another one--there's
Record #2. Another good read gives me---whoops! Wha v happened?
Record #2--Smith, J. Next record (which we think is #3 but of course
is #4)--Smith, No Well, Smith, M., is a newcomer to the family.
Ho, hum. You figure a way out of it.
2.
702 or 705 usageo Let's start with the 760 buffer containing the
five records above, and further suppose they are 200 characters each
so the control mark is in position 9990 This is the contents of the
buffer after the tape writingQ Now the internal computations in the
EDPM develop an exception item, which should be printed out on the
directly-connected HSP o This is a 120-character printout and a "write"
instruction addressed to this 760, with the printer selected, moves
this 120 characters (plus a record mark in the 702 or a group mark
in the 705) into the buffer; it goes into the first 121 positions,
wiping out most of Record #1 above, because the storage counter was
reset to 000 (automatically) after the previous tape write. Now let's
print this line. Well, let's see---on the 702 it's followed by a record
mark, and that stops the character transfer for one print line. But
on the 705 it's followed by a group mark, which doesn't. Ah--but the
printing involves a RWS 0006 instruction, which automatically changes
the group mark to a control mark---and that not only stops printing
one line, but stops printing from the buffer. But, now, wait a minute-therevs already a control mark in position 999 and a group mark in
399 (the last position of Record #2 above), plus this one in 120, which
1152.52
(2)
1152.52
(Continued)
is the one we want changed to a control mark. Precisely how does
the 760 know this? Ah, from the storage counter, of course. After
the transfer into the 760 buffer, it is set at the next character
position, so it's easy for the 760 to know that the control mark
goes into the previous character position. Really, it's quite simple
when analyzed. So that takes care of this situation very nic.ely.
And now, my fine-feathered friend, would you mind taking a look back
at the first example, after Record #2 has been moved into the buffer,
with the control counter set at the position of the first character
of Record #3 when the memory-to-buffer transfer is complete? If the
reasoning on this print operation is correct--just how Oh earth did
you get Records #3, #4 and #5 on, tape?
The second case is the more important. The 760 buffer is a non-destructive type of storage; i.e., information is replaced character by character only insofar as necessary. It's quite possible (more to the point,
a virtual certainty) to be faced with a rather short--say 400 character-record in the buffer, and have all sorts of record, group and control
marks left over from previous information in the remaining positions.
Just how does the 760 know which one to change into the control mark?
These two examples are facetious treatments of conditions which occur
commonly in practice. Their existence is not limited to on-line operations, but could arise in off-line printing. So far as can be determined, there are only two possible explanations: (1) IBM is far from
complete in its explanation of how the 760 operates; or (2) its use involves a large volume of housekeeping which is not mentioned or explained.
There definitely are inconsistencies and impossible conditions in the
methods of operation contained in the IBM information manuals on this
unit.
Revised 2/15/57
1153
1153
COMMENTS PERTAINING TO ON-LINE USE OF TYPE 760
The principal purpose of the Type 760 is, of course, to serve as a control and buffer storage unit for high-speed printer operations. In fact,
the possibility of using it as a tape control unit in CPU work is most
likely strictly a by-product of the fact that IBM, consistent with its
previous equipments, has seen fit to adapt the high-speed printers to
on-line operations under program control.
1153.1
Use of Type 760 as Tape Control Unit Only. The only justification which
can be seen for this type operation is the fact that a 760 is necessary
for a high-speed printer, and may be available during periods when the
printer is not working productivelyo Even then, it makes for a rather
expensive and slow tape control operation:
Ie
With one tape, the cost is $2,350 base rental; with two tapes, it is
equivalent to $1,450 each. This is extremely high compared with ten
tape units attached to a Type 754, averaging $750 per tape, or eight
with a Type 777 IRC, averaging $822. (These two pro-rate the cost
of the tape control unit)e
2.
The 760 reads only or writes only; it cannot do both simultaneously.
With a maximum tape record size of 999 characters, the maximum transfer rate of characters, reading from one tape and writing on another,
is 5,990 per second. This is reduced by the time necessary to unload
and load the buffer. The Type 777 can operate at twice this speed
as can the Type 754, although the latter cannot overlap tape reading/
writing and computation.
3.
The 760 buffer and memory transfer rates, at 0.023 ms per character,
is extremely slow-4this is 23 ms for a 1,000-character recorda By
way of contrast, the Type 777 unloads a similar record in 9 ms.
(And Univac II is of course still faster).
4.
In virtually all cases, tape records must be of a fixed and constant
length, and the individual items also must be of constant length.
Although a variable-length master item is possible on the 705, the
fact that neither record nor group marks are transmitted into memory
would add considerable housekeeping to a program.
These factors indicate that the only reasonably justifiable uses for the
760 as a tape control unit would involve small subsidiary files during
periods when high-speed printing was not being donee It also appears
basic that any such usage would permit handling more tape units than the
number of 754s or 777s available would permit.
1153.2
Use of Type 760 as On-Line Printer Control Unit Only. An on-line printer
is highly useful in program debugging, but Univac philosophy from the beginning has been to divorce mass-output printing from the central processor. With EDPMs of the general logic and characteristics of Univac II
and the IBM 702 and 705, there is no economic justification for on-line
printing. The closest approach to it occurs when the amount of internal
computation is just about equivalent to the time necessary to print one
line--in the case of the 760, either 60 ms or 120 ms. Such a balance is
1153.2
1153.2
(Continued)
seldom if ever achieved in practice. The net result is that the printer
is t'hung uptl waiting for the next line to be prepared, or the computer
ftstalls tt waiting for the previous· line to be printed. At the costs of
these equipments--better than $20,000 a month for the central processor
with its tape units, and from $3,200-$3,900 for the printer with its
760--it doesn't take much lost time to add up to considerable cioney. In
the case of the 705, preparation of tapes for off-line printing will always be faster using the Type 777 IRC than with a· directly-connected
HSP; the difference in buffer loading time is appreciable (0.009 ms per
character for the 777 vs. 0.023 ms for the 760).
There is little question that a directly-connected printer can be a convenience. There is no question that it is expensive-six to seven times
as costly as a tape unit.
1153.3
Use of Type 760 With Tapes and Printers On-Line. I3asically, this com.bines the remarks of the previous two sections. The new factor introduced is the characteristic of the 760 that permits printing and tape
writing of the information in the buffer storage. Unquestionably this
is necessary Or highly desirable at times. Equally unquestionably, the
magnetic tape prepared for an off-line printing operation~ exactly the
same thing. It also takes less computer time to prepare in the 702 or
705 s ys tems •
In general, there seems to be little economic justification for including
a 760 as an on-line part of a 702 or 705 installation. To the extent
that the cost of the unit reflects additional hardware necessary to use
it in this manner, it can be considered over-priced; it contains more
than is needed.
1159
1159
SALES POINTS ON THE TYPE 760 CONTROL AND STORAGE UNIT (ON-LINE OPERATIONS)
Although largely based on the previous discussion, some of these points
are also appropriate with reference to high-speed printer operations;
these are separately discussed in 41300
1159.01
Contradictory Features of the 760. Point out the examples developed in
1152.52. If nothing else, possibly an explanation of these apparent inconsistencies can be obtained.
1159.02 Character Transfer Rate Between 760 and Tapes. Because it reads only
or writes only, but not both simultaneously, the effective rate of
handling tape information with a 760 is extremely slow. If used at all,
in on-line operations, it is probably best confined to reading or writing,
but not both.
1159.03 Character Transfer Rate Between 760 and Memory. Extremely§10w at 0.023
ms per character. Because the EDPM can be doing nothing else during 760
and memory information transfers, it is tied up for an excessive period
of time. This is true on either tape or printer operations. Point out
Univac's faster buffer unloading and loading--about six times as fast as
the 760.
1159.04 Cost of Using a 760 as Tape/Printer Control Unit. Withqtwo tapes, the
per-tape cost 9f using the 760 as a tape control unit is $1,450, almost
twice as much as using a Type 777 and more than twice as much as the pertape cost of Univac (pro-rating a part of the Univac II main frame cost
to tape control and buffering). Furthermore, both the Type 777 and Univac's rI and rO tanks permit simultaneous reading and writing; this makes
the cost-of-performance ratio of a 760-connected tape more than four
times that of Univac II.
1159.05
On-Line Use of a 760 During Time Not Required for Printingo As noted in
1153.1, it may be that the printing load of an installation does not
occupy a full shift, and that the 760 may be available for some 705 or
702 work during this otherwise idle (but paid for) time. This may make
it possible to do in one run what might otherwise require two, if the
other types of tape control units in the installation are already used
to their tape unit capacity. Assuming the use of the 760 in this case
does not introduce additional rental, it sounds as if it may have merit.
For the 760: This is definitely a possibility.
Against the 760: There are at least two very definite factors working
against such part-time use:
(1) The off-line printer installation may not be in close proximity
to the main EDPM installation; it often is in a separate room.
In this case, it may be impossible or impracticable to physically
move the 760 and tape units to the computer area~ (There is a
definite maximum distance that can separate a 760 from the EDPM;
the specific distance applicable to the 760 is not known, but
most likely does not exceed 50 feet).
(2) Because the program must be written with the 760 in mind for a
specific purpose, the 760 and its tapes must always be available
for the machine run. This poses first-rate scheduling problems
1159.05
1159005
(Continued)
(particularly when the off-line printer operation has a twohour breakdown) for both printer and central processor. An additional deterring factor to this type of planning is that both
printing and computing workloads and peak periods change as time
goes on; it is extremely risky to hang a procedure on the assumption that printing will only take four hours a day next year because that's all it takes today.
1159.06
Inconsistencies Between 760 and 777. " Both types of tape control units
are equipped with the so-called "'early start" "feature. The maximum time
a buffer and memory transfer can take during the overlapping start of
tape motion is:
777
760
Writing (memory to buffer)
s:oins
9:2ms
Reading (buffer to memory)
4.6 ms
4.0 ms
Both use exactly the same tape unit. It is elsewhere explained why
the difference exists in times for reading and writing, but why are the
corresponding times different for the 760 and 777? Is it possible that
one group of IBM engineers donVt know what another has done? This inconsistency is similar in nature to that of the use of the control mark in
the 760 which was introduced as a brand new character code doing something apparently handled alright by the existing group mark in the 777.
1160
1160
IBM TYPE 766 DATA SYNCHRONIZER
The IBM Type 766 Data Synchronizer is a control and buffer storage or
synchronizer unit used in all Type 709 EDPM input-output operations except those involving information transfers between memory and magnetic
drum (Type 733) or memory and cathode ray tube (Types 740 or 780). It is
required when using magnetic tapes, or when on-line card reading/punching
or printing are involved. Strictly speaking, it is not a tape control
unit, but because it incorporates many of the functions commonly associated with such devices, it is included in this section.
The Type 766 has been announced in both Models 1 and 2. There is no
known difference in characteristics, function or price, and consequently
no distinction is made in this discussion.
1160.1
Q.Q..tl:
1161
BASIC CHARACTERISTICS OF THE TYPE 766
Purchase - $210,000
Rental
3,500 monthly
Purpose:
Synchronize and control input-output
data transfers in Type 709 system
Number associated with 709:
1 to 3 (maximum)
2 per synchronizer
Input-output channels:
1 to 8 (maximum) Type 729 Tape Units
Equipments per channel:
(with 1 Type 755 Tape Control Unit)
1 - Type 716 printer
1 - Type 711 Model 2 Card Reader
1 - Type 721 Card Punch
Magnetic core synchronizing register
Type of buffer storage:
Capacity of buffer (synchronizer): 36 bits (one 709 word)
Transfer rate, memory-synchronizer: 12 microseconds per word
Transfer rate, synchronizerVaries with type of unit
input/output units
Control features:
See 1162
Original 2/15/57
1161.1
1161.1
EQUIPMENT RELATIONSHIP OF A TYPE 766 IN A 709 SYSTEM
The place of the Type 766 as an input-output synchronizer in a Type 709
installation is depicted graphically in this illustration, which shows the
complete array of equipment types which may be associated with a single
Type 766. Up to three synchronizers, each with the same maximum array of
periphery uni ts, may be attached at one time to 'the 709 processor.
Type 709
Memory
711
Card
Reader
Type 766
Synchronizer
716
Channel
Channel
Printer
A
B
721
Card Punch
755
Tape Control Unit
755
Tape Control Unit
Any combination of units not exceeding the maximum of each individual
type may be associated. The only restriction is that a 716 printer must
be used when either a 711 card reader or 721 card punch is attached.
Each of the three Data Synchronizers which may be associated with a 709
installation can have a $ i'tilllar maximum number of input-output units, or
any lesser number. It is not necessary that each channel have the same
configuration of peripheries.
OnlV one channel in the Data Synchronizer can be used for card equipment
(711, 716 or 721): the second channel is available only for tapes. The
channels of the first 766 are identified as A and B; of the second, C and
D; of the third, E and F. By designation, A, Cfand E are used f6r-both
tapes and card equipment; B, D and F are restricted to tapes.
Original 2/15/57
1162
1162
METHOD OF OPERATION OF TYPE 766 DATA SYNCHRONIZER
The Type 766 not only provides synchronizing facilities between the 709
memory and input-output units, but also has certain control functions intended to free the CPU of much of the "clerical" instruction execution
associated with reading and writing. The nature of these control functions will become clear during the course of the explanation of this section.
The explanation of this section is divided into three phases:
(1) A general discussion of the features and methods of operation of
the Data Synchronizer;
(2) Specific information on magnetic tape uses; and
(3) Factors pertaining to the use of card equipment, by which is meant
readers, punches and printers used on-line.
1162.1
Functional Components of Type 766 Data Synchronizer Channel. Each of the
two data synchronizer channels (DSC) in a 766 contains hardware to perform some control operations, as well as a synchronizer or buffer to
match the internal 709 speed with that of the slower input-output equipmente Illustration 1162.1 on the next page depicts these functional
components graphically, and will be found useful in going through the
method of operation. The second DSC (B, D or E) is identical with the
first, except that it contains only the circuitry used with Type 729 Tape
Units and omits that associated with card equipment. In each 766, the
two DSCs are completely independent of each other in operation; each contains the features shown in the illustration.
The special registers included in each DSC and a brief statement of their
functions follows; their use is further explained in subsequent sections.
1162.11
Synchronizing Register. This is a 36-bit register which holds one word
from memory while it is being written on tape, or assembles one word (six
characters) from tape before high-speed transfer into memory. It thus
serves as a one-Word buffer between the 709 memory and tape units. Its
use is limited to tapes, and for this reason it is sometimes referred to
as the Tape Synchronizer. (AI though of no importance from a programming
viewpoint, it is probable that this register actually consists of two
36-bit halves, the input (or output) to tape alternating between them.
This technical detail is not mentioned in preliminary information on the
Type 709, but timing considerations indicate this physical configuration).
1162.12 Control Word. This is a program constant placed in the Channel by an
instruction (or automatically,as will be noted later). The control word
contains the information necessary to permit 766 operation independent of
main program instructions.
1162.13
Word Count Register -- WR. This register contains initially the number of
words to be read or written; capacity is 15 bits or a decimal equivalent
of 32,767. The contents of WR comes initially from bit positions 3 - 17
of the Control Word. (WR) is automatically reduced by one each time a
word is moved to or from memory.
Original 2/15/57
1162.1
Illustration 1162.1
REGISTERS USED IN TYPE 766 DATA SYNCHRONIZER
755
Tape
Control
Unit
709 MEMORY
Etca
YI 1
Y
M
Word Count
S 1 2 3 •
M 11
Address
. . .17
21" •
A
0
a35
Output Word #1
Output Word #2
Etc.
Address Register
Triggers-
15-Bit Address
Y 11
Location Register
LR
AR
15-Bit Word Count
Word Count Register
WR
This chart shows the makeup of the control word in 709 memory, and its
location, with the registers in the 766 into which the various elements
of the control word are distributedo The synchronizing register is also
a part of the 766.
Original 2/15/57
1162.14
1162.14 Address Register -- AR. AR contains initially the 15-bit memory address
of the. first word to be read or written; this comes from bit positions
21-35 of the Control Word. (AR) is automatically increased by one each
time a word is moved to or from memory.
1162.15
Location Register -- LR. LR contains the location of the current control
word plus one; i.e., if more than one control word is required for an input/output operation, it contains the address of the next control word.
The loading of this 15-bit register is an automatic part of moving a
Control Word into the DSC; its contents are set to the location from
which it was taken plus one.
1162.16 Channel Control Indicator A (Trigger A). The bit in the "s" (sign) position of the Control Word is stored in this trigger. The general interpretation of the bit is:
o - Last control word in an input/output sequence;
1 - More control words following the one being usedn
1162017
End-of-Record Control Indicator B (Trigger B). The bit in the "Itt position of the Control Word is stored in this trigger8 A "1" in this bit
position forces a change in a Control Word whenever an end-of-record condition is reach in tape operation; what happens is explained more fully
in later sectionsQ
1162.18
Trigger CQ The bit in the "2".position of the Control Word is placed in
this trigger. A "1" in this bit position causes a change in the normal
sequence of obtaining new Control Words; this likewise is explained more
fully in subsequent sectionso
1162.19
Summary of Control Word Decomposition in DSCs. The storage of the various
positions of the 36-bit Control Word in the DSC occurs as a part of its
transfer from memory to the DSC. The bit position distribution is summarized below:
Bit S
1
2
3 - 17
18-20
21-35
'l +1
To Trigger A
To Trigger B
To Trigger C
To Word Count Register (WR)
Not used
To Address Register (AR)
To Location Register (LR), where Y is location of
Control Word being transferred into DSC.
The 709 is provided with an instruction which permits storing this channel
information in the main memory. The contents of the several registers
in the DSC are reassembled into the same Control Word positions, except
that (LR) is transferred into bit positions 3-17 and the contents of WR
do not appear in the reconstructed word.
Original 2/15/57
1162.2
1162.2
Basic Method of Input/Output Operation with Type 766. The explanation
first outlines the fundamental operating method of the Type 766, with
ramifications introduced beginning in 1162.30 For simplicity, tape writing
is generally used in the examples; reading is understood to be essentially
the converse of writing except where note is made of differences. Notations conform to those in Illustration 116201.
1162.21 Instructions Initiating Output (Input) Operations. Data transfer operations are initiated by a sequence of two instructions not necessarily
consecutive in a program). The first of these designates the type of
transfer (input or output), the Data Synchronizer Channel and the specific
unit involved. The second places a Control Word in the DSC registers and
initiates the actual transfer of information.
Write (Read) Select Instruction. A "Read Select" or "Write Select" operation code prepares the 709 for the appropriate action with the unit specified by the address part of the instructiono The address includes three
factors: (1) The DSC involved; (2) the particular unit (tape or card
equipment); and (3) in the case of tapes, the mode of writing or reading
(straight binary or character code)o The interpretation of the address
is in groups of three bits (octal), as follows:
Bits 21-23 -- Always zeroes and not used.
n
24-26 -- Selects DSC Channel involved:
1 -- Channel A (Tapes and card equipment)
" B (Tapes only)
2
3
" c (Tapes and card equipment)
" D (Tapes only)
4
5
" E (Tapes and card equipment)
"
F (Tapes only)
6 --.
11
27-29 -- Distinguishes tape (from card equipment:
2 -- Tape
3 -- Card equipment
t1
30-32 -- (1) If tape is indicated by bits 27-29, designates mode
of writing:
o or 1 -- Character Coded Decimal
2 or 3 -- Binary
(2) If card equipment is designated, selects unit:
2 -- Card reader
4 -- Card punch
6 -- Printer
n
33-35 -- (1) If tape is indicated, the specific one of eight
maximum is designated (1 through 0, octal, the
carry being reflected into bit positions 30-32).
(2) If card equipment is indicated, this group is always a "1".
Thus, a "Read Select 02207" (octal notation) alerts tape unit #7 of DSC
B to prepare to read in character code fashion; "Write Select 03361"
alerts the printer attached to Channel Co
Loading the Control Word. The loading of the control word is carried out
by executing a "Reset and Load Channel" instruction contained in the main
program. This instruction must be executed within 3 ms after alerting
a specific DSC for a read operation, or 7 ms for a writeo The channel
Original 2/15/57
1162.21
1162.21
(Continued)
involved is indicated as a part of the operation code itself; thus there
are six separate "Reset and Load Channel" instructions in the 709 repertoire. (This is probably necessary, because several "Read (Write) Select"
instructions may be encountered before anyone of them is followed by a
"Reset and Load Channel;" thus the Channel must be repeated).
Assume DSC A is selected. Execution of this instruction transfers the
control word, whose location is given as the address part of the "Reset
and Load Channel At' instruction, into the DSC A registers, the storage
of the various parts of the Control Word being accomplished as depicted
in Illustration 1162.10 The Control Word (CW) contains the memory address
of the first word to be wri tten, the numbeor of words to be written, and
possible bit entries in position S, 1 and 2 of the word. These are moved
into the Address Register, the Word Count Register and Triggers A, Band
Co Simultaneously, the location of the Control Word plus one (i.e., the
address of the "Reset and Load Channel An plus one) is stored in the
Location Registero All of these registers are a part of DSC A and are
separate from the registers of the 709 itselfo Upon completion of executing
this instruction, the 709 central processor is free to continue with other
computation; from this point on, the Data Synchronizer Channel circuitry
takes over the handling of the input or output operationo
1162.22 DSC Functions in Output (Input) Operations. As soon as the Control Word
has been broken up into the various DSC registers, one access (12 microseconds) is made to the memory location indicated in the Address Register
and the word found there is transferred to the (Tape) Synchronizing Register. Simultaneously, (AR) is increased by one and (WR) decreased by one.
The 36-bit word contained in the synchronizer is now written onto the
selected output tape, each group of six bits being written in parallel
with a check bit automatically added¢ Thus one word becomes six character
codes on tape; about 400 microseconds are required for this operation at
67 microseconds per character.
When the synchronizer is empty, the DSC inspects the Word Count Register.
If this is not zero, more words are to be written and the location of the
next output word is contained in the Address Register. At this point,
the DSC interrupts the computing operations of the 709 central processor
to transfer this one word into the synchronizer. In general, this is an
interruption of computing; the DSC "steals" one cycle to make the transfer
of the required wordo It may occur at any point in a normal 709 instruction; for example, the interruption may come after the first cycle of a
two-cycle (24 microsecond) instruction. It always occurs at the end of
709 cycle and always tak~one cycle to completeo This causes a 12~micro
second delay in the 709, but does not change any of the registers used
in the main processor--all of the necessary information is stored in the
DSC registers. The reason for interrupting, rather than overlapping,
709 computing is that most of the 24- and 36-microsecond instructions in
the 709 require a memory access on each cycle and only one access is possible per cycle. Thus control circuitry is probably simplified by considering all these short-time instructions as needing accesses each cycle,
even though such operations as register shifts require only an instruction
access to main memory.
Original 2/15/57
1162.22
1162.22
(Continued)
Graphically, the timing of the interruption may be depicted quite easily;
the normal uninterrupted execution of two successive instructions is
shown first, followed by two possible timings of the DSC interruption:
Normal Instruction Execution
12
24
Instruction Operand
Access & Access &
Interpreta- Execution
tion
36
R
Q
P
Instruction
Access &
Interpretation
First Instruction
48
(Time in microseconds)
S
Operand
Access 8;
Execution
Second Instruction
DSC Interruption Between Instructions
0
I
12
24
P
Q
First Instruction
ITra~~~er J
Word to DSC
From Memory
DSC Interruption Within Instruction
0
12
24
P
.Q
DSC .,.
I
First - -
I- -
36
60
S
I
Second Instruction
3,
t
- -Instruction
- - DSC Transfer
48
R
48
R
S
6f
Second Instruction
Let us immediately answer the rather obvious question which now arises:
nWhat time does this save--it appears to sandwich in the transfer of
information to and from the DSC synchronizer by delaying execution of instructions, adding to total computing tfmeo" The answer is that it doesnVt
save any time in the transfer of data between memory and the buffer--but
what difference does it make, overall, if 1,000 words are transferred in
one chunk requiring, for example, 12 ms, or in 1,000 separate chunks of
12 microseconds? The total time is the sameo What is saved is rather
expensive buffer hadware--a one-word synchronizer costs a lot less than
a 1,000 buffer (which is nothing but another name for synchronizer)o The
concept used in the 709-766 is several years old, but this is one of the
first equipments announced which will incorporate the featureo As a minor
note, it is quite probable that we have seen the last of the large-scale
EDPMs with big buffers; they very likely will be limited to equipments
now in production or so far along in the design stage that the introduction of newer techniques cannot be incorporated without a major overhaul
of the equipment logico Although the specific method used in the 709-766
is different, a similar approach is incorporated in the design logic of
LARCwh'ich we are now buildingo
Original 2/15/57
1162.22
(2)
1162.22
(Continued)
There is one class of instructions in the 709 which permits the memory access required by the DSC to actually overlap internal computation; these
are the multiply and divide operations, which take from 2 to 20 cycles,
during most of which no memory access is required. If a DSC needs a new
word from memory while the 709 is engaged in executing one of these instructions, the transfer of the word will occur during the same cycle in
which part of the computation is performed. In business data processing,
multiplication and division instructions require only a very small part
of total computing time and this feature is of little significance; in
scientific computations, however, these instructions may take from 15-40%
of total computing time, and some reduction in total processing time may
be achieved by this simUltaneous DSC-memory transfer.
Now back to the synchronizer operation. The write-out of information on
tape continues word by word, one memory aCcess being captured by the DSC
each time the synchronizer is emptied--in other words, about once every
400 microsecondso Each time the contents of thw Word Count Register is
inspected and reduced by one, until ultimately it reaches zero. At this
point, the fundamental method of the writing operation is terminated; the
check character is added to the tape record, the unit is disconnected from
the Data Synchronizer and the tape comes to a stop. The Data Synchronizer
is available for some other input operation, or output, as soon as the
check character has been written.
Reading tape records of known length is basically the converse of writing.
The end of the tape record, signified by the end-of-record gap, and the
reduction of the Word Count Register to zero occur simultaneously, and the
input operation is completed with the transfer of the last word into the
709 memory.
With the basic method of operation thoroughly understood, we are now in a
position to go into some of the variations possible.
1162.3
Use of Multiple Control Words. If, when the Word Count Register is reduced to zero, Trigger A is "on"-i.eo, a ttl" was contained in the "s"
position of the Control Word-- the writing operation is not discontinued,
but instead the DSC obtains a new Control Word from memory and continues
writing, using the contents of the new Control Word to determine the next
memory word required and the number of words to be written. In other
words, successive Control Words can be loaded, as required, in the DSC,
each designating an initial memory address and word count. New Control
Words are obtained from successively higher memory locations; it will be
recalled that the Location Register holds the address of the current Control Word plus one--i.eo, the address of the next higher memory location.
So long as the "s" position of a Control Word is a binary "1," the DSC
automatically secures the next successive Control Word each time the
Word Counter is reduced to zero. There is no limit on the number of
Control Words which can be used; theoretically, it is possible to write
a complete tape solidly from beginning to end using a large number of
Control Words. Uses of this feature are discussed beginning in 1163.
Thus, the writing operation continues until the current Control Word has
a "0" in the "s" position, which means the A Trigger is set to "off. f1
Original 2/15/57
1162.3
116203
(Continued)
In reading from tapes, if Trigger A is "on," end-of-record gaps are ignored and the tape continues reading until the Word Count Register is
reduced to zero. Thus several tape records may be read successively and
transferred into memory as one larger grouped reoord. This permits passing over the gaps at full speed in 10 ms, rather than requiring the longer
time involved if the tape stops after reading each record.
1162.4 Reading/Writing Under Trigger B Control. A binary "Itt in bit position 1
of a Control Word places Trigger B "onu and the input-output operation is
under what IBM calls "record control." Because the sequence of events is
somewhat different for input as compared with output, the two types of
operation are discussed individuallyo
1162041 Reading Under Trigger B (Record) Control. Under Trigger B control, when
an end of record (inter-record gap) is detected on tape, the next successive Control Word (determined by the address in the Location Register)
is transferred into the DSC regardless of the status of the Word Count
Register or of Trigger A, and the next following record is read from the
tape. Trigger B thus overrides Trigger Ao Uses of this feature are discussed in subsequent sections.
1162 .. 42 Writing Under Trigger B (Record) Control. If Trigger B is on during a
writing operation, the status of Trigger A must be taken into account.
When the Word Count Register is reduced to zero, the next action depends
on the status of the A Trigger:
(1) If the A Trigger is "on," (i.eo, a ttlt1 in the"Stl position of the
Control Word), an inter-record gap is forced on tape and the next
Control Word in sequence is moved into the DSC. Writing continues
from this point on the basis of addresses and word counts in the
new Control Word.
(2) If Trigger A is "off" (ioe., a "On in the "s" position of the Control Word), the output operation has been completed, and the DSC
disconnects the tape unit involved.
Thus a single t~eset and Load Channel" instruction executed by the 709
central processor may initiate the reading or writing of a large number
of tape records without further intervention by the main program. So long
as Control Words contain a "1" in bit position 1, Trigger B is turned "on"
and one more record is read or written; this cycle continues until the
Control Word of the last tape record involved is moved into the DSC; the
"0" in bit position 1 turns off the B Trigger and the transfer reverts to
the fundamental method of 1162.2.
116205
Altering the Sequence of Transferring Control Words. Normally, successive
Control Words are obtained sequentially from memory locations indicated
by the address in the Location Register. There are two methods of
altering the normal sequence of obtaining Control Words, one automatic
and one under program control.
1162.51 Altering Sequence by Trigger Co If in a sequence of Control Words, one
contains the bits "001" in the S, 1 and 2 positions (Triggers A and B
Original 2/15/57
1162.51
1162.51
(Continued)
off~ C on)~ the normal use of the Control Word does not apply.
Instead
of being taken as the next Control Word, the address portion only (bit
positions 21-35) is used as the location of the next Control Word; this
address is placed in the Location Register and an access made to that
memory location for a new Control Word. Succeeding Control Words, if
called for, are obtained in the normal manner beginning with this new
location. Thus, when Trigger C is on, the normal location of the next
Control Word is used only to obtain the address of the one actually to
be used; this is virtually identical to the "indirect addressing" used
in the 709 central processor.
.
1162.52 Altering Sequence by Main Program Instruction. The second method of
changing the normal sequence of obtaining Control Words is through use
of a "Load Channel" instruction in the main programo The initial "Reset
and Load Channel" can seldom be used, because it is executed immediately,
regardless of the status of the DSC operation which may be going on.
The "Load Channel" instruction differs in that it does not take effect
until completing the action required by a Control Word which has a binary
"In in bit position 2 and also in position Sand/or l--in other words,
Trigger C and Trigger A and/or B set to "on." With this bit combination,
the DSC will do one of two things:
(1) Execute a "Load Channel" instruction provided one is waiting; or
(2) Disconnect the selected input/output unit and terminate the operation if no t~oad Channel" instruction is waiting.
In either event, the action specified by the Control Word containing the
necessary bit combinations in the S, 1 and 2 positions is completed before the "Load Channel" is executed or input/output terminatedo Use of
this program-interrupt feature requires synchronizing the input/output
routine with the main program because the "Load Channel" instruction
must be available at the time the DSC completes the action of a Control
Word having the specified bit combinationso
116206
Status of Data Synchronizer Registers. It may be necessary, during the
course of a main program routine, to know the status of a specific inputoutput operation previously initiatedo This is accomplished by a "Store
Channel" instruction, which transfers the contents of the several DSC
registers into a memory location indicated by the address portion of the
instruction. The reconstructed word is similar in format to a Control
Word, except that the contents of the Location Register is inserted in
the normal Word Count bit positions, and the Word Count Register is not
sent into the reconstructed wordCl The status of the three Triggers, of
the Address Register and the Location Register are transmittedo This
instruction interrupts, but does not terminate or alter, the DSC operation. The instruction execution is interlocked to prevent the transfer
of the register information during the time a new Control Word may be
in process of movement, or during a change of the contents of the Word
Count and Location Registerso The normal input/output sequence being
carried out by the DSC continues without changeo
Original 2/15/57
1162.7
1162.7
Testing Status of Data Synchronizer Channels. Two conditional transfer
instructions are included in the 709 repertoire to test the condition of
a selected DSC. These are "Transfer if Channel A in Operation" and
"Transfer if Channel A Not in Operation," with similar instructions for
the other five channelso A channel is considered to be in operation if
it is not ready to execute a new input/output instruction.
116208 Check Indicators and Verifying Accuracy of Data Transfers. A number of
special indicators are provided for each of the channels to permit program testing of various conditions applicable to input-output units and
of the accuracy of the data transfer. These are:
(1) End-of-tape tests; (one in each DSC and in each tape unit)
(2) Beginning-of-tape tests;
(3) Redundancy indicator, turned on when a check bit or check character
error occurs;
(4) End-of-file indicator (turned on by a tape mark).
The redundancy indicator is operative whether or not the contents of a
tape record are being transmitted to memoryo It should be noted that it
is not necessary to read into memory an entire tape record; a dummy control Word can be inserted when it is discovered that a particular record
is not required, eliminating the transfer of the contents of th synchronizing register; the tape~ however, continues moving until an interrecord gap is reached and the normal testing of tape characters is operative during this periodo
116209
Interaction of Triggers A, Band C; Word Count; Tape Recordso The various
actions taken by a DSC after transmitting each successive word of information are determined by the setting of the three Triggers and of the Word
Count Register; in addition, in tape or card reading, the end of record
(card) also must be considered. The various possibilities are summarized
in the tables on the next page, which groups the combinations causing each
type of DSC operation. In these tables~ "W" is the contents of the Word
Count Register, "0" being zero and "¢" being non-zeroo These tables are
included for reference purposes only; the basic methods of operation of
the DSC with magnetic tapes have been outlined in the previous paragraphs
of this section; specific comments applicable to card equipment are contained inl163.
Original 2/15/57
1162.9
(1)
Table 1162.9 (1)
Tape and Card Reading Operation of DSC
Triggers
End
of
000
100
No
No
No
No
No
No
No
1
0
1
010
o 1 1
110
III
100
100
0
0
o 1 0
o 1 0
¢
0
110
0
110
0
0
¢
001
001
001
001
o
o
101
o
o
1
o
o
1
~
1
1
1
1
¢
III
III
¢
101
o
III
o
o
000
101
101
Continue transmitting words from 1/0 unit into memory
Bring in next Control Word from location specified
by Location Register
No
Yes
Bring Control Word into DSC from location specified in
address part of this Control Word
yes}
Yes
Yes
Yes
Yes
If "Load Channel" instruction is waiting, execute it.
If not, disconnect input/output unit.
NO }
No
If "Load Chann'el It instruction is waiting, execute it. If
not, disconnect unit but move it until it reaches end-ofrecord
Yes~
Yes
Ignore end of record and continue transmitting words
from 1/0 unit into memory
Yes'
000
o
o
010
o
Read tape without transmitting words to storage. At end
of record, bring in Control Word from address specified
by Location Register.
o
o
Read tape without transmitting words to storage. At end·
of record, if "Load Channel'" instruction is waiting,
execute it. If not, disconnect unito
000
1
1
No
Disconnect unit from 709, but move it to end-of-record.
Yes
Disconnect unit from 709
"¢" means "not zero"
1162.9
( 2)
Table 1162.9 (2)
Tape and Card Writing Operation of DSC
Triggers
000
100
101
010
011
110
111
Continue transmitting words from memory to I/O unit
o·
110
Bring new Control Word into DSC from location
specified by contents of Location Register
001
001
Bring Control Word into DSC from location specified
in address part of this Control Word
1
0
000
o
101
o}
010
O}
011
111
o}
o}
Write end-of-record designation and disconnect unit
If load channel instruction is waiting, execute it.
If not, write end=of-record. and disconnect unit
Write end-of=record indication and bring in new
Control Word
Write end-of~record indication. If ·~oad Channel"
instruction is waiting, execute it. If not, disconnect unit
If "Load Channel" instruction is waiting, execute it.
If not, disconnect unit
1163
1163
METHOD OF OPERATION OF TYPE 766 DATA SYNCHRONIZER (Continued)
Section 1162 outlined the basic characteristics of the Type 766 DSC and
the general method of operation. Some of the comments pertained only to
magnetic tape units, because some types of operations are peculiar to
that type of input/output device. This section discusses in more detail
the use of the DSC in operations concerned with both magnetic tapes and
card equipment. The last part of this section has some sumITary comments
on the use of DSCs in the 709 system.
116301 Modes of Reading/Wri ting With Magnet:ic Tapes
Section 1162.2 describes
the basic method of operation of the DSC in reading or writing, and these
general comments apply fully to magnetic tapes. The basic additional note
to be covered is the observation that the 709, like the 704, is capable
of two separate modes of writing on magnetic tape: Binary or charactercoded decimale Which method is to be used in a given operation is determined by the address of the instruction selecting the tape unit; each
tape unit has two logical addresses, one for each method of writing or
reading.
0
If information is to be written on tape in binary recording, the successive words of 36 bits each are recorded as six groups of six bits on the
tape. To each group of six bits an odd check bit is automa~ically added;
thus each word from memroy becomes six separate character codes on tape,
each group of six bits and the added check bit being written successively.
There is no change in the six data bits recorded; they are simply transferred to the tape. Reading is the converse of writing,the six bits from
each character code on tape being transferred without change into the
synchronizing register. The check bit is not included in this assembly
of successive characters on tape, but is checked to determine validity of
reading and a check error arises if the bit is invalid.
If information is to be written on tape in character-coded form, the successive groups of six bits in the words moved into the synchronizing
register are not written as they appear, but are automatically converted
from the internal 709 code to the 702/705 character codes. (Refer to
5152 for a table of the internal and external character codes used in the
704/709 systems)o In this case, an ~ check bit is added to each group
of six bits after the conversion; the resulting recording is identical
with that of the 702/705 systemso It should be noted that information in
binary form cannot be converted to a character-coded form, but must be
written on tape in binary. The reason is fairly evident--there are only
48 valid character codes of si~ bits each in the internal language, although 64 are possible; the ,16'/ invalid bit combinations could not be converted and most likely woul-d (/'gIve rise1b,Gf~qin input/output error. Reading
is the converse of writing--the tape characters are automatically converted
into the internal codes and assembled into the synchronizing register befor transfer of each word into memory.
The "READ" instruction must specify whether binary or coded characters
are to be read; because of the check bit differences it is obvious that
the equipment must know in advance which type of recording is being used
as input. Although there appears to be no reason mixed types of recording
Original 3/1/57
1163.1
1163.1
(Continued)
could not be combined on a single output tape, the difficulties of providing the proper read instruction for each successive tape record precludes this as a practicable method. Anyone tape reel will be in
binary, or in character-coded decimal, but not both.
It should be observed, however, that anything in the 709 can be written
as binary information, even though the information actually represents
character codeso The distinction of data as being binary or cha~acter
coded is purely human; the 709 doesnVt know the difference and handles
everything internally (with a few exceptions) as if it were binary. In
use, character-coded magnetic tape will be used for off-line printing,
punching and card reading functions, or for preparation of tapes to be
used as input into 650, 702 or 705 installations. Tapes, including
master files, which are to be used as later input into the 709 itself
will almost always be written in the binary mode.
1163.2
Use of Tvpe 711, Model 2, Card Reader With Type 766. One channel of a
DSC may have attached to it, under program control, one Type 711, Model
2, card readere It is also required that a Type 716 Printer be attached
in order to use the card reader. Only one card reader can be associated
with each Data Synchronizer, using only Channels A, C or E. Thus the
maximum in one 709 installation is three.
The basic method of card reading is described in section 3110 and is not
changed in 709 usage. 72 columns of the card are read and converted into
24 "card image words," each word having a bit combination corresponding
to the presence or absence of holes in 36 columns of one row of the card.
These 24 words are read sequentially and transferred into memory, through
the synchronizing register, according to data in the control wordo The
principle of the transfer is identical with that of reading from magnetic
tapes. The basic difference is in the timing of the movement of the
successive word images; the DSC automatically adjusts itself to the timing
of the card reader. This is explained in 3110.
It is important to bear in mind that the information transferred into
memory is in card image form and thup is neither binary nor character
coded (unless the card read is pun9hed in binary fashion as discussed in
3110; this is almost exclusively peculiar to scientific input and does
not occur in punch card business input). A rather extensive conversion
progranisnecessary to change the 24-word card image into either straight
binary or character coded form for internal 709 processingo It does not
appear that the special "conversion" instructions incorporated into the
709 are particularly applicable in this case; if this is true, a long and
time-consuming program will be necessary to effect the conversiono
1163~3
Use of Type 716 Printer with Type 766. One channel of a DSC may have attached to it, under program control, one Type 716, Modell, printer. Only
one can be associated with each Data Synchronizer, using only Channels
A, C or E. The,maximum in one 709 installation is threeo
The basic method of printing is described in 4110 and is not changed in
Original 3/1/57
1163 .. 3
1163.3 (Continued)
709 usage, and is essentially the reverse of the card reading mode summarized in 1163020 Each word transferred from 709 memory through the DSC
to the printer is a 36-bit representation of 36 columns of one card row;
thus 24 words must be transferred to permit printing a 72-character line.
Information maintained in either binary or character-coded form in the
709 must be program-converted to the proper punch card word imageso As
in the card reader, the DSC automatically adjusts the transfer of words
from memory to the timing necessary for the operation of the printer.
1163.4
Use of Type 721 Card Punch With Type 7660 One channel of a DSC may have
attached to it, under program control~ one Type 721, Model l~ card punch.
It is also required that a Type 716 Printer be attached in order to use
the puncho Only one card punch can be associated with each Data Synchronizer, using only Channels A, C or Eo The maximum in one 709 installation is three ..
The basic method of card punching is identical with that of printing, and
is discussed in section 4510. As in printing, the DSC automatically adjusts the transfer of words from memory to the timing necessary for the
operation of the puncho A maximum of 72 columns of a card can be punched.
116305
Interruption of Data Transferso Each Data Synchronizer Channel is capable
of handling one--and only one--data transfer at a timeo That is, if it
is writing on one tape, it cannot simultaneously read a punch card, or
read from another tapeo Once an input/output unit has been selected, and
electronically connected to the DSC, it commences the required operation
and continues with that operation until one of these two conditions occurs: (1) Failure of the DSC to receive the first control word within
a specified time limit after the unit has been selected--the time varies
depending upon the type of unit involved; or (2) completion of the specified transfer as indicated by the word count and status of Triggers A, B
and C held in the DSCo
In some applications, however, "real time" data of an intermittent nature
may be used as input to a 709; such information usually requires immediate
processing, which means interrupting whatever the EDPM is doing at the
moment to process the urgent datao The only means of input to the 709
is through a Data Synchronizer Channel; where real time information is involved, it may then be necessary to interrupt an input or output operation to permit receipt of these datao The exact method of doing this is
not detailed in available information on the 709, but the method of accomplishment is this: A control or key word with a specified bit combination is fed into the 709 (in a manner not described) where a matching
bit pattern is retainedo Receipt of this key word automatically transfers
program control to memory location 0004 and simultaneously stores the
location of the next instruction in the interrupted sequence in location
00020 If the DSC through which the urgent data is to be received is busy
in some input or output operation, the transfer is stopped by a "Reset
and Load Channel" instruction which has a zero word count and a "0" bit
for Trigger Ao This instruction would be included in the sequence beginning in location 0004; the entry of a new Control Word into the DSC is
forced by the instruction, and the unit involved in the transfer is disOriginal 3/1/57
1163.5
1163.5
(Continued)
connected, freeing the DSC for the "real timet! information. It should be
noted that entry of the key word into the 709 does not of itself stop a
DSC transfer; this is done by programming included as a part of the subroutine beginning in location 00040 The actual stoppage is accomplished
by the "Reset and Load Channel t ' instruction, which places a new Control
Word into the DSC registers.
Rather evidently, a "Reset and Load Channel 1t instruction which would stop
input/output action could be executed at any time in a program for any
reason, including whims of the programmer. An example of its use in other
than ureal time" instances is in the reading of a fairly long string of"
tape records with a sequence of Control Words. A tape reading error could
occur early in the string and it would then be desirable to terminate the
reading to permit a corrective re-read subroutine to be entered. This
would be done by inserting several instructions testing the status of the
error indicators at periodic intervals--say every 20 ms. If an error is
detected, the program transfers to a corrective subroutine which includes
a "Reset and Load Channel tt instruction to stop the reading immediately,
rather than cycling through the remaining tape records specified by the
normal sequence of Control Words. The use of this instruction, then, is
a third means of stopping the transfer of data through a DSC.
1163.6 Reason for the Number of Data Synchronizer Channels. The 709 can be
equipped with up to three Data Synchronizers, ora maximum of six channels.
This limitation in number does not appear to be based upon any processing
considerations--ioe., there is no indication by IBM that six is the optimum or maximum number of inputs and outputs which can be handled by the
709 internally--but is a limit imposed because of technical considerations.
The reason is this, using writing on tape as an example. When the 36 bits
in the synchronizer have been written on tape, the next word must be obtained from 709 memory and moved into the synchronizer and this must be
done by the time the next character is to be written on tape o The transfer of the word takes 12 microsecondsa But now let us look at the worst
possible condition. Suppose the last word written on tape reduced the
word count to zero, but did not terminate the writing sequence. This
means that another Control Word must be moved into the DSCG Further, it
might be an indirectly-addressed Control Word. This means that it is possible, in the worst case, to require an access for the indirectly-addressed
word, another access for the new Control Word and, finally, an access for
the information word to be written on tape. This total of three accesses
requires 36 microseconds. With six DSCs, it is possible--again in the
worst case--that each of the six requires exactly the same treatment. In
the worst case, a total of 18 memory accesses, requiring 216 microseconds,
might be required. With a nominal 400 microseconds tape time for handling
one word, and the engineering requirement for a good margin of tolerance
on this figure--an insistence on something like 50% is customary--it is
apparent that this "worst case tt is an engineering limitation. It is
necessary to have sufficient time to take care of the worst possible occurrences, and addition of one more Data Synchronizer, which would add 72
microseconds to the time requirement, would add to more than can be considered safe.
Original 3/1/57
1163.6
1163.6 (Continued)
This summarization also makes it evident that the ttsynchronizing register"
is not a 36-bit affair, but rather consists of two 36-bit halves which
alternate back and forth. With only one 36-bit register, it would be
necessary to obtain the next word within 67 microseconds, less some tolerance; this is the nominal time between successive tape characters. It is
obvious that 6 x 12 is 72, and with single registers it wouldnOt even be
possible to move six new data words into the DSCs within one character
time, let alone provide for the possibility of new Control Words. This
is a purely technical consideration; the programmer doesn't need concern
himself with this facto
Original 3/1/57
1164
1164
INPUT/OUTPUT TECHNIQUES POSSIBLE WITH TYPE 766 DATA SYNCHRONIZER
The control circuitry included as a part of each channel of a data synchronizer makes possible a number of input-output techniques which, heretofore, have been beyond the scope of EDPM capabilities. One of the basic
purposes of the unit is to free the central processor of much of the socalled t'clerical tI effort associated with input-output operations. There
is no question that the DSC logic incorporates some highly desirable
features, although some complexities have been introduced and not all
characteristics are without disadvantages.
At this time, there is no available information on the various inputoutput operations and techniques which IBM will advocate as promoting the
most efficient use of the 709 system. Consequently this section is not
exhaustive, but it does include a number of possibilities.which are immediately obvious from the characteristics of the DSC and some comments
on their general applicability. The following discussion should therefore
be considered only as preliminary; as time permits, more detailed analysis
will be made and the information contained herein supplemented or revised
as necessary.
1164.1 Elimination of Internal Memory Transfers. ~robably the most significant
single characteristic typifying the 709 EDPM system is a definite effort
by IBM to virtually eliminate multi-word internal memory transfers. Three
characteristics of the equipment shriek this conclusion:
(1) Absence of any form of multi-word memory transfer instruction;
(2) The t'indirect addressing t• scheme incorporated into both the 709
central processor and DSCs; and
(3) Accomplishment of input-output operations by the use of multiple
Control Words, if desired.
The attempt to circumvent the normal requirements for internal transfers
is,lof course, aimed essentially at freeing the central processor from
a time-consuming clerical chore. As such, it is not concerned directly
with the Data Synchronizers. Nonetheless, it is the inclusion of the
multiple Control Word feature that completes the logic necessary to attempt the scheme. In short, a fundamental objective of the 709 system is
to eliminate multi-word internal memory transfers and to substitute in
lieu thereof single-word transfers of words which become, ultimately, DSC
Control Words.
The internal 709 characteristics which are intended to accomplish this
objective are discussed in section 5150. The possibilities and techniques
included in this section are concerned largely with DSC use, and mention
the 709 only to the extent necessary to understand the basic methods of
operation possible.
1164.2 Sequencing of Words as Part of Output Operation. Through the use of
multiple Control Words, each of which can transfer the source of output
information to a new memory location, it is possible to assemble individual
records into an index-number sequence as a part of the output operation,
rather than by internal memory transfers. This likewise can be done as a
part of input to the 709, but the utility of the technique is not so obvious as it is in output. There are two major areas of interest in EDPM
Original 2/15/57
1164.2
1164.2
(Continued)
applications--sort-merge functions and the reassembly of master file
information to incorporate additions and deletions into the file.
1164.21 Sort-Merge Technique. Sorting and merging in the 709 will consist not
of sequencing the individual records by means of multi-word memory transfers, but rather of arranging the locations (i.e., memory addresses) of
the index words into a sequence such'lhat the individual records will be
in proper order if they are taken in the succession of location words.
This is accomplished by placing the addresses of the index words into a
series of other words and, through "indirect addressing,tt sorting these
location words into the desired sequence. (The internal use of "indirect
addressing" and amplification of this technique are contained in the discussion of the 709 central processor, 5150). As a final fillip, these
location words become Control Words for output operations, and the actual
sequencing of the complete individual records does not occur until they
are written on tape. Thus the internal memory transfers normally associated with a sort-merge operation reduce to transfers of one-word
"location words," which are so set up in format that they are directly
usable as DSC output Control Words.
For example, the record size may be such that a thousand of them can be
accommodated comfortably in memory, which would require the use of 1,000
Location-Control Words. Initially, these would all have a "binary""l" in
the S position; after the sequencing of the Control Words and prior to
initiating the writing operation, a tlO" would be placed in the 1,000-th
word. The first word is placed into the applicable DSC with a tfReset and
Load ChanneP' instruction, after which the ttl fI in the S posi tion (which
turns on Trigger A) causes a new Control Word to be moved into the DSC
every time the Word Count is reduced to zero (i.e., every time a complete
record has been transferred to tape). The "Off in the S position of the
last Control Word stops the write-out, and the entire 1,000 records have
been placed on tape in proper sequenceo. The same basic principle can be
used for either sorting or merging, although the details vary depending
upon the type involved.
Insofar as the storage on tape of a sequenced string of records is concerned, it may be noted that there is no particular reason for them to be
all of the same number of words. Each Control Word has its own Word
Count, and these can be different in successive Control Words. The
writing operation, then, can easily accomodate records of varying lengths.
Reading them back in for subsequent merging passes adds a few difficulties,
requiring some additional internal programming, but presents no insurmountable barriers. Again, refer to the 709 central processor discussion
for further details. This brief explanation can be concluded with one additional statement: The sorting of variable-length records may well be
economic in the 709-766; it certainly is always possible.
1164.22 Master File Tape Block Reassemblyo The conventional techniques of providing for additions and deletions within a master file call for the reassembly of an input block into parts of two output blocks by means of
multi-word internal memory transfers. Not only does the 709 not have such
an instruction, but it is quite evident that the technique to be advocated
Original 2/15/57
1164 .. 22
1164.22
(Continued)
will consist of treating Control Words to accomplish additions and deletions of information. This certainly can be done--the 709 has a number
of ttbit manipulation" instructions which facilitate the creation of such
Control Words, although at this time the optimum approach is far from
clear. One possible method is outlined briefly in the next paragraph,
not for the reason that it is of necessity particularly apropos, but
because it illustrates a use of the ttindirectly addressed" Control Word,
(See 1162.51), a feature whose application is not too obvious.
Suppose a master file is maintained in tape blocks of 1,000 words maximum
size.. Now establish 20 output Control Words, each moving 50 data words
from memory to the tape. Imagine also that the fourth group of words (in
other words, in the 151-200 range) has a 10-10 word deletion.. This could be
handled in this fashion: Replace the fourth Control Word with one con'",;taiiling the Trigger C control and an address which is beginning of a
"Control Word subroutine .. " The subroutine Control Words are created to
write out, say, 15 words by the first one and 25 by the second, the address
of the first starting the writeout at the beginning of the 50-word group
and the second after the lO-word deletion. These two new Control Words are
followed by another under Trigger C control which returns to the main sequence of Control Word instructions. This appears to be a quite logical
technique; further, it tends to corroborate the belief that the indirectly
addressed Control Word is of most utility as a means of initiating a subsequence altering the normal course of a read or write operation ..
1164.3
Multiple-Block Reading and Writingo The maximum effective transfer rate
of information between tapes and memory is achieved with long tape blocks,
which minimize the occurrence of inter-record gaps and the time necessary
to pass over them
Extremely large tape blocks require, of course, considerable memory space if the entire block is to be heldo However, the
percentage increase in effective transfer'rates drops rapidly as blocks
become longer. For example 9 a 500-word 709 block results in an effective
tape speed of about 95.5% of maximum; increasing the block to 1,000 words
boosts this to about 97 e6%e The effective transfer rates for the two tape
block sizes are about 149300 and 14,600, respectively, compared with the
15,000 instantaneous rateo Rather evidently, any increase above 1,000
words (6,000 tape characters occupying 30 inches) gives only a marginal
improvement in tape transfer rates ..
0
Several factors indicate that shorter tape blocks are also a common occurrence--the requirements of periphery units is a good example, and the
incidence of tape read/write errors also mitigates against extremely long
blocks. In some instances, file format considerations,' particularly with
variable length records, may also tend to shorten the number of inches of
tape packed with characters before a gap is reached.. In these cases--and
they are probably going to be quite common--some tape time can be saved
if the gap can be covered at full speed, rather than having the tape stop
'for each one.
Break-up of an output into several records or reading of multiple blocks
on input axe both possible with the Data Synchronizer; the settings of
the three triggers and the status of the Word Count Register which make
Original 2/15/57
1164.3
1164.3
(Continued)
this possible are shown in Tables 1162.9 (1) and (2). This facility has
the effect of permitting multiple blocks to be read and written with only
one initiating action by the central processor and, secondly, of giving
a somewhat higher effective transfer rate by covering the inter-record
gaps at full tape speed~
Some techniques known to be in use on the Type 704 EDPM indicate that
the following possibility of handling variable-length master file records
may be advocated by IBM~ The basic premise of the technique is that
records vary not only in the number of detail elements which may be appended to them, but also may have a significant percentage of "zero" entries within the elements which exist. For example, a stock control file
may have words for on hand quantities, back orders, due ins and total demands for the past three months" For many of the stock items, some or all
of these may be zero. Comparable conditions exist in many types of business filesa The method of approach is this:
One word is set up as a "control word" for 36 other possible ones, a "I"
in a bit position indicating that the corresponding word is present and
a "ott, that it is noto The only words which are included in the file are
those corresponding to "Iso" If a word has a non-significant entry, it
simply does not appear at all~ Thus, if a control word has at its first
ten bits a "'1110000101,'· only five words of information follow; the bits
in the control word are used to identify their nature& With this technique, individual records within one ~ile can, of course, vary widely in
the number of words of information associated. In general, considerable
input-output time can be saved by handling records in this variable-length
form.
Suppose the maximum siz,e of any given item is 100 709 words, and that
memory space for ten such items is available. This makes it possible to
read or write ten at a time, but it would be highly convenient to know
where they were in memory; considerable computer time can be spent in
tlbootstrappingt~ through variable length records
0
First of all, let's write out ten such recordsa The ten Control Words
each contains as Word Count the actual number of words in the corresponding record, and each record begins (assume) in a memory location ending
in "00," and of course the memory addresses in the Control Words are each
100 higher than the preceding one" With a bit combination fl010" in the
S, 1 and 2 positions of the Control Words 9 this permits writing out each
of the ten records as an individual record on tape--i&eo, a gap is tlforced"
between each one.. Because each Control Word has the actual word count of
the item, only the significant words of information are written out~
The read-in is likewise accomplished with the uOIO·9 bi t combination in
all Control Words, each with a Word Count entry of "lOOtt. (The maximum
individual record size)o The memory addresses in these Control Words are
also the successively higher '000" locations
Wi th this type of control
(Trigger B or end-of-record), either reduction of the word count to zero
or presence of a gap on tape causes a new Control Word to be loaded into
the DSCo In other words, each record is read into a location beginning
0
Original 2/15/57
1164 .. 3
(2)
1164 .. 3
(Continued)
at a memory address ending in "00", and continuing in successively higher
addresses until then inter-record gap is reached on tapeo The next tape
record is stored beginning at the next higher HOO" addresso Thus~ regardless of individual lengths, the beginning memory address of each
record is in a known location and, at the same time, the tape time necessary for transferring "pad characters" or zero entry words is eliminated.
These features of the DSC make available for the first time an IBM data
processor which is considered flexible enough to handle the variablelength records which comprise the vast majority of business files.
1164.4
Passing Over Unwanted Information. In some processes, it is necessary
to read over large volumes of information to secure a relatively small
amount needed for actual processingo If the file does not require updating (that ii~ if it does not need rewriting), considerable time may be
saved internally in the EDPM if the transfer of unneeded records into
memory can be avoided. Usually, this requires that the first fewcharacters or words of each tape block be read and tested, but if this can be
accomplished quickly, it may be possible simply to pass over most of the
block without actually transferring it into memory_
Although there are other methods of accomplishing essentially the same
thing, the logic of the 709 requires that two conditions be met before
the objective can be realized:
(1)
(2)
It is necessary to know when the required number of words to be
tested have actually been read in from the tape and transferred
to memory; and
The equipment must be capable of suppressing the transfer into
memory at any point along the tape blocke
Both of these requirements are fairly obvious.
In the 709-766, the first is met by the "Store Channel" instruction (see
116206). This permits a program to obtain access to the status of the
various DSC registers and~ by program testing, to determine if the required number of words has been transferred into memoryo If they have,
it is known that new information is in the low-address end of the input
area and the necessary tests can be made to determine if the item is required for the computing process. If it is not, the program can execute
an immediate "Reset and Store Channel" instruction, with a zero word
count entry and a bit in the 1 position to turn Trigger B on" This stops
the transfer of information into memory, but the tape continues moving
to the end of the record" This takes care of the second requirement
listed above.
1164.5
Continuous Reading/Writing" The nature of the memory accesses for data
transfers to and from the DSCs makes it evident that there is no necessity of waiting for a complete block to be transferred into memory
before writing it out. As a matter of fact~ it is perfectly possible for
one DSC to access a specific memory location on one cycle and another DSC
access it on the next cycle. Provided necessary computing can be done in
Original 2/15/57
1164~5
(Continued)
the time available, there is no logical reason to prevent an input tape
packed solidly from beginning to end (no gaps) being read in, processed
and written out on an equally-solid tapeo For example, the input can be
read into 1,000 words of memory, the Control Words transferring the input
back to the beginning location at the end of every 1000th wordo Output
can be on the same cycle, for the same memory area, lagging the input by,
say, 500 wordso The only operating requirement is that of keeping the
input and output in reasonable synchronism; tape unit variations mea~
that the reading rate will differ, normally, from the average bya slight
amount. Thus it is possible for reading to be slightly faster than the
average character rate, which means that the input might eventually "catch
up" with the output; or, conversely, the reading may be slower, with the
result that the output would overtak.e the input.. The occasional insertion
of tVStore Channel tt instructions would permit checking the lag between output and input and 9 if required, taking some corrective action ..
As stated above~ the idea is purely idealistic.. The accuracy of tape
reading would dictate some much shorter tape block size, and the increase
in effective transfer rates is so minute with blocks of more than 1,000
709 words that longer ones give no significant gains in speed and may have
offsetting disadvantages in terms of accuracY0 These are, however, practical considerations; there is certainly nothing in the DSC logic that
makes them impossible.
1180
Index (1)
IBM Type 777 Tape Record Coordinator
1180
IBM Type 777 Tape Record Coordinator (General statement)
1180.1
Month1 y Rental
1181
Characteristics of the Type 777 IRC
1182
Operating Characteristics of the type 777 IRC
1182.1
'IRC-Tape Transfers (Operating Mode A)
1182.11
Tape to mc (AI)
1182.12
'IRC to Tape (A2)
1182.13
IRC to Tape and Tape to IRe (Read While Writing) (A3)
1182.14
Comments on Operating Mode A
1182.2
Memory-'IRC-Tape Transfers (Operating Mode B)
1182021
Memory-to-IRC-to-Tape With Early Tape Start (B1)
1182.22
IRC-to-Memory and Tape-to-IRC with Early Tape Start (B2)
1182 .. 23
Memory-to-IRC-to-Tape and Tape-to-IRC (B3)
1182.24
Comments on Operating Mode B
1182.3
Memory-IRC Transfers (Operating Mode C)
118204
flMicrosecond Interrogation fl (Operating Mode D)
1182.5
By-Pass Coordinator (Operating Mode E)
1182.6
Checking in the IRC
1182 .. 7
Testing Status of IRe
1182.8
Summary of mc Operating Times and Modes
1183
Length of Tape Records and IRe Operations
1183 .. 1
Effect of Length of Grouped Records of Constant Length
118302
Effect of Variable Lengths of Grouped Records
1183.3
File Maintenance Using the Type 777 mc
1184
File Maintenance with the Type 777--Records Passing Through Memory
1184 .. 1
Maintenance of Files With Fixed Master Item Length
1184011
Factors Pertaining to 705-777 File Processing
1184012
Factors Pertaining to Univac II File Processing
1184013
Comparison of Univac II and IBM 705-777
1184 .. 2
Maintenance of Files With Variable Length Master Items, Fixed Tape Blocks
1184.21
Factors Pertaining to 705-777 Processing
1184.22
Factors Pertaining to Univac II Processing
1184.23
Comparison of Univac II and IBM 705-777
1184 .. 3
File Maintenance With Variable Length Tape Records
1184.31
Factors Pertaining to 705-777 Processing
1184.32
Factors Pertaining to Univac II Processing
1184.33
Comparison of Univac II and IBM 705-777
1185
,File Maintenance Using ttMicrosecond Interrogation"
1185 .. 1
File Maintenance Facts Inherent in t~icrosecond Interrogation"
1185.2
Level of Activity in File Maintained by "Microsecond Interrogation"
118503
Determination of Processing Time Savings
1185.31
Comparison with File Maintenance Using Type 754
1185.4
Possible Duplication of Information in an'Interrogated File
Original 1/31/57
1180
Index (2)
INDEX (Page 2)
1186
Use of Two or More Type 777 IRCs in the IBM 705
1186.1
Implications of ·'Doub1e Buffer" Operation Using Two lRCs
1186.2
Factors to be Considered in a Double-Buffer Process
1186 .. 21
IRC Control Sequences
1186.22
Testing of Master File Items
1186.23
Input-Output Areas
1186.24
Number of Tape Units
1186.25
Multiplexing ,of Processing
1186.26
Machine Runs Before and After a Double-Buffer Process
1186.3
1tMicrosecond Interrogation t ' Processing With Two lRCs
118604
Double-Buffered "Read Into Memory" Processing
1186.41
Implications of Block Reassembly
1186 042
Testing of Master File Items
1186.43
Constro1 Sequence
1186 .. 44
Productive Processing
1186 5
Mixed "Microsecond Interrogation" and "Read Into Memorytt
6
1187
Systems Factors Affecting Double-Buffered Processing
1187.1
Tape Unites and Number of Input-Output Files
Necessary Duplication of Tape Units
118'011
1187 .. 12
Other Inputs and Outputs
1187 .. 2
Duplication in Memory
1187 .. 3
Non-Overlapping Operations
1187 31
Machine and Data Transfer Errors
1187 .. 32
Check Points
1187 .. 33
Typewriter Printouts
1187.34
Machine Failure
1187.4
Summary of Double-Buffer Processing Speedup
1187 .. 5
Multiplexing of Two Different Jobs
0
1188
Comparison of Univac II With Doub1e-IRC Type 705
118801
Costs of Typical EDPM Installations
1188 .. 2
Comparable Installations, Univac II and IBM 705
1188021
Two IRCs With One Input and One Output
1188 .. 22
Two IRCs with Multiplexing
1188.3
Cost of Performance, Single-Buffer Type Operation
1188s4
Cost of Performance, Double-Buffer Type Operation
1188041
"-Microsecond Interrogationo 1t
1188.42
ttReading Into Memory·'
1188 .. 5
Summary of Comparative Costs
1189
Sales Points on Type 777 Tape Record Coordinator
Original 1/31/57
1180
1180
IBM TYPE 777 TAPE RECORD COORDINATOR (TAPE CONTROL
UNIT~
The Type 777 Tape Record Coordinator (usually referred to as "TRC tl ) is
a tape control unit used to connect Type 727 magnetic tape units to the
705 central processor and to provide buffer storage facilities to permit internal computation to proceed during tape read/write operations.
The 705 can be equipped with up to ten TRCs, or up to ten total TRCs, 754s,
760s and 774s in any combination. (This statement may be" incorrect; the
addressing structure is such that ten TRCs in addition to ten other types
of tape control units could be used. A definitive answer is of no immediate importance, because no known or projected 705 installation involves
anything close to ten such tape control units for on-line operation).
Because IBM is lending heavy sales emphasis on the use of two or more
TRCs to provide a "double-buffered" 705, the method of use and the systems implications of the 777 will be examined in considerable detail.
1180.1
Monthly Rental:
$3,000.
1181
CHARACTERISTICS OF THE TYPE 777 TRC
Number of Tape Units Controlled:
Type of Buffer Storage:
Character Capacity of Buffer:
Number of Buffers:
Transfer Rate, Tape-Buffer:
Transfer Rate, Memory-Buffer:
Tape Unit Addressing Structure:
Checking of Information Transfers:
1 to 8 (maximum) (Type 727)
Magnetic Core
1,024 7-bit characters; 1,022
usable positions
1
0.067 ms per character
0.009 ms per character
06XO to 06X3, 06X5 to 06X8, where:
"06" designates TRC use;
'~" is 0 to 9 and designates
specific TRC (of ten maximum)
Unit digit is specific tape unit
attached to TRC.
06X4 and 06X9 have special meanings, discussed in 1182
(1) Check bit for each character
(2) Check character for each record
The characteristics of the various modes of operation of which the TRC
is capable are explained in detail in 1182.
1182
1182
OPERATING CHARACTERISTICS OF THE TYPE 777 TRC
The'TRC is capable of five separate modes of operation in a 705 installation; these are designated as A, B, C, D, E. The applicable mode is
determined by instruction sequences in a 705 program or by the address
of the instruction selecting the TRC. The discussion which follows assumes an understanding of the tape format used in the 705 (1103.6). The
explanations all use tape unit addresses in the range 0600 through 0609;
they are thus associated with the TRC designated by the ten 9 s digit "Ott.
1182.1 TRC-TAPE TRANSFERS (OPERATING MODE A)
This mode of operation involves transfers of data between the buffer
storage of the TRC and tapes; memory is not involved. The maximum record
which can be handled is 1,022 characters. There are three sub-modes of
operation.
1182 ... 11 Tape to TRC (AI) .. A grouped tape record can be read from anyone of the
eight tape units into the buffer; the sequence of instructions (using
0602 as the tape unit from which reading is to be done) is:
SELECT
0602
CONTROL (RTS) 0016
Selects tape unit
Initiates read only operation
The first instruction alerts the proper tape unit; the'RTS 0016 initiates
the actual transfer. Each requires 0.051 ms of internal computing time;
the 705 is free to continue with other computations during the actual
transfer from tape to the buffer.. The time required for the transfer is
10,LOo067N ms, where N is the number of characters in the record read in.
The characters of the grouped tape record are read into the buffer beginning at its lowest position and continuing until all characters hav~
been read. Recognition of the inter-record gap on tape causes a group
mark to be placed into the next higher buffer position and the check
character is placed immediately after the group mark.. Necessity for inserting the check character and group mark limits the tape record to a
maximum of 1,022 characters.
1182.12 TRC to Tape (A2). A grouped record in the TRC buffer can be written on
anyone of the eight tapes, without reading from a,nother, by this sequence of instructions, again using 0602 as the tapeuni t.:
SELECT
0602
CONTROL (WST) 0017
After executing the WST 0011 instructions, the 705 main frame is free
to continue with otheroperaiions.. Actual writing requires 10
O.067N
ms, whereN is the number of characters in the buffer record. The writing operation continues character by character until the group mark is
reached in the buffer; it is not placed on tape but generates the interrecord gap and terminates the writing operation.
f
1182.13 TRC to Tape and....Iape to TRC(Read While Writing) (A3). It is possible to
write a grouped record from the TRC buffer onto dne tape while simultaneously reading a record frqm another tape into the buffer. Using 0601 as
the read tape and 0602 as the write tape, the sequence of instruction to
accomplish this is:
1182.13
(1)
1182.13
(Continued)
SELECT
0601
CON1ROL(PRW) 0015
Select the tape for reading
Prepare to read while writing (RWW) --
SELECT
CON1ROL
Select the tape for writing
Write from IRC onto 0602 and read from 0601
into mc buffer
0602
0017
The "read t• tape is always the first selected. Once these instructions
have been executed (0.204 ms), the 705 central processor is free to continue with other computations while the actual data transfers occur.
This sub-mode of operation is one of the three principal ones which IBM
is using to promote the use of t'double buffers t• on the 705 with two (or
more) IRCs.
Because the IRC has only a single buffer, it is necessary to assure that
the record being read in does not destroy the information to be written
outo This is accomplished by automatically delaying the beginning of
the read operation for 7 ms o This chart shows the timing of the two
operations:
Writing
10 ms
Time to
Start Tape
7 ms
Delay
Pending
Write (00067 x Noo of Characters)
10 ms
Time to
Start Tape
Read (00067 x No. of
<
~
Cha~ac~~r_s_)_________
)
Time, measured from completion of CIRL 0017 instruction
01
~--------~~~
The 7ms is sufficient to assure that the reading in of information will
involve buffer positions already written out; it will be noted that the
lag is such that the reading is about 105 character positions after the
writing
Thus the writing from and reading into the buffer occur simultaneousl y, the reading ·'lagging·' the writing by 7 ms, or about 105 characters. If Nl is the number of characters to be written, and N2 the
number to be read, the total time required for the operation is the
greater of
10 I 00067 Nl
and 17 I 00067 N2
RWW, using a IRC, places certain restrictions on the two tape units involved. The maximum of eight which can be associated with a IRC are
divided into two groups, depending upon the units digit of the tape address:
0
Group I
0600
0601
0605
0606
Group II
0602
0603
0607
0608
1182.13
(2)
1182.13 (Continued)
In mode A3,
is in Group
restriction
one set can
one tape must come from each
I, the write tape must be in
applies only to each pair of
be Read 0600--Write 0602 and
group; i.e., if the read tape
Group II, or vice versa. The
read-write operations; that is,
the next Read 0603--Write 0605.
1182.14 Comments on Qperating Mode A. Modes Al and A2 do not appear to be of
major importance. Although on occasion they will be convenient and necessary, either operation can often be combined with some other type of
data transfer using other TRC modes. The "HWW", sub-mode, however, is
always involved in an operation using the so-called "microsecond interrogation" feature of the TRC (see 1182.4), and is fundamental in the use
of this feature.
l182.2MEMORY-TRC-TAPE TRANSFERS (OPERATING MODE B).
This mode involves the transfer of information between the 705 memory and,
the TRe buffer, together with either a tape reading or writing operation.
In each of the three sub-modes, the memory-TRC transfer requires O.009ms
per character (9 ms for 1,000 characters), plus instruction execution time,
during which period nothing else can be done in the 705.
1182.21 Memory-to-TRC-to-Tape With Early Tape Start (Bl). Because the time to start
a tape in motion and accelerate it to full speed is 10 ms, and the time -necessary to load 1,022 characters (the maximum possible) from memory into the
TRC buffer is 9.2 ms, it is possible to start an output tape in motion at
the same time the transfer from memory to the TRC buffer begins. By the
time the tape has reached writing speed, the buffer will be fully loaded.
The sequence of instructions, which does not overlap other instructions
using the TRC is:
SELECT
Write
0603
(m)
Select output tape unit 3
Write record from memory into TRC and from
TRC to tape unit 3, with early tape start
As \lsual, the group mark following the last information character or record
mark in memory terminates thememory...;to-TRC transfer.. Likewise, the group
mark in the TRC terminates the tape writing operation.
1182.22 TRC-te-Memory and Tape-to-IRC With Early Tape Start (B2). The converse
of Bl is to transfer the contents of the TRC buffer into memory and to
read an input record into the buffer, using early tape start for the input. Unlike writing--which can be electronically timed to begin exactly
10 ms after the execution of thetfwrite" instruction--readingmust begin
when the first character is reached; because of mechanical variations in
stopping' the tape, it is possible that the first character will be reached
before 10 ms have elapsedo With 9&2 IDS required to transfer a full buffer
load into memory, it is therefore possible to reach the first tape character
before the me has been emptiedo To provide a sl,lfficient margin of safety
the execution of the tape1Jlovement in this sub-mode recognizes two conditions.
(1)
If the record in the TRC buffer is 512 characters or less (maximum
transfer time to memory is then 4.6 ms). the acceleration of the
1182.22
1182.22
(Continued)
read tape occurs simultaneously with the beginning of the TRC-tomemory transfer
(2)If the record is longer than 512 characters, the tape start is delayed until after the 512th character has been moved into memory;
i.e o, it is delayed for 4 6 ms.
0
0
No program attention is required to achieve the delay if necessary; the
tthardwaret~ checks the number of characters in the buffer and inserts the
delay automatically. The sequence of instruction in both cases is:
SELECT
READ
0605
(m)
Select input tape 5
Transfer the contents of the IRC buffer into
memory, then read tape 5 into the IRC with
early tape starto
The IRe buffer contents are transferred into memory, character .by character, until the group mark is reached in the TRC. This terminates the
transfer, but the group mark itself is not transferred into memory; the
last character transferred is the one preceding it in the buffer. The
705 can do nothing else during the IRC-to-memory transfer, at 0.009 ms
per character, but is free to continue other computations during the
tape reading operation. If N is the total number of characters in the
tape record to be read, the total time until the reading has been completed is either 10 I 00067N or 1406 I 0 .. 067N ms, depending upon whether
the initial buffer contents exceeded 512 characters ..
1182 .. 23 Memory-to-TRC-to-Tape and Tape-to-IRC (B3) .. This sub-mode is quite
similar to A3, except that the output record is transferred from memory
to the IRC buffer as a part of the instruction sequence, rather than
being in the buffer alreadyo The sequence of instructions is similar
to that of 1182.13:
SELECT
*
*
*
*
PRW
SELECT
WRITE
0601
0015
0602
(m)
Select the input tape, 1
Prepare to read while.writing
Select the output tape, 2
Transfer the record beginning in (m) into the
IRC buffer, then write it on tape unit 2, beginning the movement of the tape at the same
time the transfer into the IRC commences. Then
read one record from tape unit 1 into the IRC
buffer, delaying the start of the "read" tape
for 7 ms after the start of the 'twri te tt
0
The restriction on tape unit number assignments discussed in 1182013
applies here also. Otherwise, the method of execution is identical with
the individual transfers previously discussed. The 705 central processor
is tied up during instruction execution and the transfer of the output
record from memory to the IRC at 00009 ms per character. The tape write:
operation requires 10 I 0.067 Nl ms, and the read 17 100067 N2 , where Nl
and N2 are the number of characters in the output and input records.
Revised 2/15/57
1182.23
1182.23
Mode B3 is a combination of mode Bl and the tape reading portion of mode
B2. The dumping of the output record into the IRC and writing it, with
early tape start, is identical with mode Bl. The start of the read tape
is delayed for an arbitrary 7 ms. for the same reason as the delay in
the 'tread while writing" mode A3, section 1182 .. 13. The timing of the complete cycle is the same as that shown in the chart of 1182.13, the loading
of the IRC with the output record occurring during the 10 ms start time
of the output tape.
*
*
*
*
*
*
*
*
*
*
*
*
(Continued)
1182.24 Corrunents on Operating Mode Bo. Sub-mode B3 is the most important, because
it is the one normally used in file maintenance operations. In execution,
it is the IRC operation corresponding most closely to the 5n (write) and
3n (read) orders ina Univac. The chief difference is that Univac has
separate input and output buffers, and no delay in the read tapes is
necessary, as it is in the 705-777. The transfer rate between memory and
buffer is also considerably faster in Univac II--approximately 2 75 ms
for a full block of 720 characters, compared with 60>48 ms for a comparable
block in the 705; the main frame is tied up for a considerably longer
period in the 705. One factor which will be referred to frequently in
subsequent discussions evaluating the use of the IRC should also be noted:
The group mark is never transmitted from the IRC to memory, but it must
be at the end of the record to terminate a memory-to-IRC transfer. The
tteCirly tape start" is likewise nothing new; all it means is that loading
or dumping of a buffer overlaps some of the tape start time, which Univac
has been doing since 1951.
0
1182.3
MEMORY-IRC IRANSFERS (OPERATING MODE C)
In mode C, data are transferred from memory to the IRC buffer, or vice
versa, with no tape action involved. This mode of operation is designated
by a 1~41t in the units position of the address selecting the tape unit;
it will be recalled that 9"4" and "9" are not valid tape unit numbers. The
form of instruction sequence for either reading (IRC to memory) or writing
(memory to IRC) is:
SELECT
0604
READ (WRITE) (m)
Select the IRC; the .'4" indicates mode C
Read (write)beginning at memory address m
The transfer of information is terminated by the group mark in memory
(writing) or the IRC (reading). On reading, the group mark is not transferred into memory, but on writing it is placed in the IRC. The time
required for the transfer in either direction is 00009 ms per character,
plus instruction execution time, during which period no other computing
can be done.
1182.4
"MICROSECOND INTERROGATION" (OPERATING MODE D)
The 777 is equipped with a feature, called ttmicrosecond interrogration,fI
which permits transferring from the buffer into memory a control field of
a few characters, rather than the entire recordo By testing an index num~
ber in the control field, it can be determined if the record is required
for processing, and it is not brought into memory unless it is. The
time saved is the O.OO9ms per character required to bring in the entire
recordo The length of the control field transferred can be of 8, 16,
Revised 2/15/57
1182./';
1182.4
(Continued)
32 or 64 characters; a switch on the 777 panel, manually set, determines
the number. Operating mode D is indicated by a "9" (not a valid tape unit
address) in the units position of the instruction selecting the TRC; e.g.,
"SELECT 0609 t " followed by "READ (m)" transfers 8 (16, 32, 64) characters
into memory beginning at address m. In use, "microsecond interrogation"
is coupled with the "RWW tf submode A3 to handle the tapes.
Mode D is being advocated by IBM for the maintenance of files which have a low
acti vi ty rate. It is the last of three principal modes of _operation of the
~
1182.5 BY-PASS COORDINATOR (OPERATING MODE E)
In this final mode of operation, the buffer storage of the TRC is by-passed,
and the record is read in from tape directly to memory, or the converse.
In this mode, its use is identical with that of the Type 754 Tape Control
Unit (1140); because nothing else can be done during the transfer of information. mode E normally would be used only for handling records longer than
1,022 characters. Unlike the 754, however, "read while writing" cannot be
done in this mode; either one, but not both together. The E mode of operation is indicated by this sequence of instructions:
SELECT
0601
CONTROL (BPC) 0018
READ (WRITE)
(m)
Select tape unit, 1
By-pass TRC
Read (write) record beginning in memory
location m
The "CONTROL 0018", instruction is used only for this one purpose, and
causes the buffer storage unit to be by-passed only for the succeeding
read or write instruction. The TRC reverts to normal status at the completion of the instruction. Although the buffer is not used, the TRC is
tied up during the entire transfer of information 10 ;. 0.067N ms, as well
as the central processor.
1182.6 CHECKING IN THE TRC
All transfers of information using the TRe, whether from tape-to-buffer
or buffer-to-memory, and all combinations thereof, are given the normal
IBM validity tests. Each characte-r is inspected for a proper check bit,
and a new check character is created and tested against that following the
record (TRC-tape transfers only; the check character does not appear in
memory). Errors turn on check indicators as follows:
(1)
(2)
(3)
Errors made in memory-to-TRC or TRC-tomemory transfers turn on
the read-write check indicator (address 0902).
Errors made in a Tape-to-TRC transfer turn on the read-write check
indicator (0902).
Errors made in a TRe-to-Tape tra'nsfer turn on the record check
indicator (0903).
The signal indicating an error in a memory-TRC transfer is available
I
I
1182.6
1182.6
(Continued)
immediately upon completion of the information transfer; the read-write
check indicator can be set either to automatic stop or to program control
(normally done), and the signal program-tested by a conditional transfer
instruction.
The signals in a TRC-tape transfer are not available until the following
read or write instruction addressed to the TRe. Again, this is normally
under program control rather than automatic stop. Because it is highly
desirable to know that a transfer of information has been done correctly,
it appears that almost all 705 users will introduce a program convention
of a "fictitious" write instruction to determine the validity of a previous read/write operation. This will consist of selecting any tape unit
and using a "write" instruction whose address is a group mark location in
memory Because the group mark is not written out, nothing will be written
on the tape selected, no tape movement will occur, and the only effect will
be to make the error signal, if one exists, availableo
0
It should also be noted that a memory-to-TRC error will frequently turn on
the machine check indicator (0901) in addition to the read-write check indicator; the corrective subroutines must provide for turning both of them
off.
1182.7 TESTING STATUS OF TRe
The possibility of connecting more than one TRe to the 705 makes it desirable to test the status of any which may be ih use; if a TRe is
engaged in some previously-initiated operation, it may be that the 705
can be used for some other computation.. To enable testing the status of
a TRe, the normal e9:Transfer Signal" instruction of the 705 is divided into
two parts when one or more 7775 are connected ..
(1)
(2)
If the address of the "Transfer Signal tt is coded for accumulator
00, the normal meaning of the instruction applies ..
If the address is coded for accumulators 01 to 15, it becomes a
"Transfer Readyn instruction for the selected TRe, the transfer
being effected if the TRe is ready to carry out more instructions;
i .. e., if it has completed a previous input or output operation ..
Through use of the "T:ransfer Ready" instruction, the status of any TRC
may be tested periodically in a program, and some other type of computation performed until it has completed the data transfers specified
previously in a program.. The use of "TI'ansfer Signal, n uTransfer Ready'"
and "Transfer Any" instructions can be done in any of the modes in which
the TRC operateso
1182.8
1182.8
SUMMARY OF IRC OPERATING TIMES AND MODES
For reference purposes, the operating times and modes of the Type 777 IRC
are summarized in the following table. All times are in milliseconds.
In the column on IRC times, the letter "Ntl designates the number of characters to be read or written when only a single tape operation is involved. If both reading and writing are involved, ttNltt designates the
number of characters written and ttN2t1 the number read.
Unit
Instruc'n Mode Selected
GIRL 0015
Prepaxe
0.051
Data Flow
Tape Time
CPU Time
None
GIRL 0016
Al
Any Tape
0.051
10 / 0.067N
Tape to IRC
CIRL 0017
A2 Any Tape
0.051
10 / 0.067N
IRC to Tape
CIRL 0017
after
CIRL 0015
0.051
10 / Oo067N l
ox
17 / 0.067N2
IRC to Tape
A3
~ITE
Bl
Any Tape
0.051/0.009N
10 / 0.067N
READ
B2 Any Tape
O.051/0.009N
10 I 0.067N - O.OO9N
14.6/0.067N - 0.009N
WRITE
after
CIRL 0015
B3
Any Tape
O.051/0.009N
10 / 0.067N l
or
17 /0.067N2
Mem. to me
to Tape
Tape to IRC
~EAD
Cl
06x4
0.051/0.009N
0.051 / 0.009N
IRC to Mem.
WRITE
C2
06x4
0.051/0.009N 0.051
D
06x9
Oo051/0.009N
READ
Any Tape
0.051
I
I
Tape to IRe
Mem. to IRC
to Tace
Tape to IRC
0.009N
Mem. to IRe
0.009N
me to Mem.
~EAD
preceded by E Any Tape
GIRL 0018
'WRITE
preceded by E Any Tape
CIRL 0018
lo/o.067N
10 / 0.067N
lo/0.067N
10
I
0.067N
Tape to Mem.
Mem. to Tape
Notes:
Modes A3 and B3: Use 10ngex IRG time.
Mode B2: Use second IRC time if N equals or is greater than 512.
Mode D:
N is 8, 16, 32 ox 64, depending on switch setting.
Mode E:
IRC time and tape time are identical.
Revised 1/31/57
1183
LENGTH OF TAPE RECORDS AND TRC OPERATIONS
The three most important operating modes of the Type 777 TRC are A3 and
B3 (read while writing) and D (microsecond interrogation). In the typical file maintenance operation, an u~dated block of the master file is
written out at the same time a block from the old file is read in; these
are accomplished most rapidly in modes A3 and B3. Periodically, it may
be necessary to either read or write a single block--for example, to read
a new block of transaction items to be processed against the master file
or to write out a block of transaction information. These may not occur
at points in the program where a simultaneous read and write is possible.
In general, however, Univac experience is that more than 90% of read-write
operations occur in pairs.
Because the TRC has only a single buffer, used for both input and output,
it is tied up until both operations are completed, in an A3 or B3 mode, or
until a single operation is completed in the other modes.. In combined
reading-writing if the reading takes longer than writing, the TRC is not
available for further data transfers until the reading has been completed;
the converse is also true. With reading lagging writing by 7 ms in a combined operation, the two types of transfers would end at the same time if
the record read in were about 105 characters shorter than that written out.
In practice, of course, it is impossible to continue very long with all input re~ords 105 characters shorter than the output. The most efficient use
of the TRC, in terms of TRC-tape transfer operations, occurs when both input
and output records are the same length. In this case, the combined operation requires 17/0.067N ms, where N is the number of characters in each
record. This corresponds exactly to the principle of UNIVAC blocks, which
are always exactly 720 characters; the only difference is that the length
of the tape record in the 705 can be any number of characters up to 1,022
maximum.
1183.1 Effect of Length of Grouped Records of Constant Length. In order to minimize the number of inter-record gaps on a tape--each of which adds 17 ms
to total tape time--it is desirable to group individual records into longer
blocks, or grouped records in IBM terminology, on tape; this is standard IBM
practice in its recommendations.. The effect of this grouping on tape speed,
in characters per second, is evident in the following table, which gives the
net speed for tape grouped records of varying lengths; each allows 0075 ms
for non-overlapping computations necessary to use the TRC (this is about
the minimum time which will be possible; frequently it will be more than
this) ..
Number of Characters
in Tape Record
100
200
300
500
750
1,000
1,022
Net Tape Speed
Characters per Second
4,100
6,450
7,950
9,750
11,000
11,800
11,900
1183.1
(Continued)
It is evident that tape records 750 characters long or longer are desirable to permit passing over unneeded master file information at the highest possible rate of speed. Likewise, it should be noted that the fastest
tape speed possible, using the TRC buffer, is only 83% of the rate of
Univac II. At best, the 705 with a single TRC takes 20% longer to pass a
file than does Univac II. It is equally important to note that the charac~rates given in the previous table do not include getting the information
into memory, but only the time necessary to pass the file through the TRC
!2.1!ti~.
The Univac II rate of 14,300 characters per second of course, is
based upon the tape data going through memory.
1183.2
Effect of \(ariableLengths of Grouped Records. If successive grouped records
on a Type 727 magnetic tape are not all of the same length, the determination of tape reading/writing time becomes much more complicated. The following table is indicative of what occurs, and will be used to explore the implications of variable length tape records in more detail. The table is
based upon simultaneous reading and writing, using the conventional times
for Mode A3 operation; the grouped record read in on one line becomes the
record written out on the next ..
Number of Characters
Tape Time (Milliseconds)
Write
Read
(1000)
1000A
900 B
600 C
850 D
100 E
1000 F
800 G
900 H
1000 I
100 J
1000 A
900 B
600 C
850 D
100 E
1000 F
800 G
900 H
1000 I
100 J
( 300)
Write
77.0
77.0
70 .. 3
50.2
67 .. 95
16 .. 7
77.0
63.6
7003
77 .. 0
1607
Total times ..
Read
Net
84.0
77.3
57 . 2
74.95
23.7
84.0
70.6
77.3
84.0
23 .. 7
37.7
84.0
77.3
70 .. 3
74.95
67 .. 95
84.0
77.0
77.3
84.0
77.0
~
811.5
The average length of the ten grouped records identified as A through J
is 725 characters.. If the tape consisted of constant records of this
length, the time to write one out and re.ad one in would be 65.6 ms; wi th
the variability shown above the average time is 73.8 ms, or about 12i-%
longer. The reason for the greater time is obvious by noting that the
two short tape records, E and J, have no effect on net TRC time; the
longer records on either side override.
Some further considerations arise from more detailed study of the
0
The input records begin in 1000, 1200, 1400, etc. with the record marks
terminating each one in 1199, 1399, 1599, etc. A group terminating the
input area may be program-inserted into 2000; it is not mandatory but
if used for any purpose must be placed there by the program--it is not
inserted into memory during a me buffer to memory transfer. The output
area is similarly set up; in this case the group mark in 4000 is necessary and must be program-inserted at the beginning of the machine run.
It does not need to be re-inserted after that ..
Now suppose that previous additions and deletions to the file have resulted in the condition that the last three records of one input become
the first three of an outputo This means that Records #1 and #2 of the
next input from tape must be added to the end to fill out the output.
Omitting test necessary for determining if any input item must be processed, it remains to consider the use of the high-speed memory transfer
in effecting the re-assembly. There are two characteristics of this
instruction in the 705 that are pertinent: (1) The data transferred must
begin in a memory location ending in a ttott or a "5; tt (2) The transfer is
terminated by a record mark in the final position of a group of five characters, i.e 0, in a memory address ending in ·°4" or "90 t. The file layout
above meets this requirement.
It will be noted that the nature of the internal memory transfer of the
705 permits only one record ata time to be movedo Since each grouped
1184.11
(2)
1184.11
(Continued)
record consists of five individual items or records, five separate
memory transfers are required for a complete reassembly of the input
block; in this example, two of them move· Records #1 and #2 to the end
(Records #4 and #5) of an output block, and the remaining three move
#3 through #5 to the first of the next output. There are two basic
methods of doing this--by "loops" or straight-line coding--and numerous
variations of each. In fact, it is somewhat pointless to develop a
sample program for this example, because any particular file considered
may well have characteristics permitting a faster program to be devised.
Therefore, only some general comments on the grouped record reassembly
are noted.
(1) The ftlooptt method usually takes fewer instructions, overall, than
straight-line coding, but generally takes more computer time. It
may be the preferred method if memory space is tight ..
(2) Straight-line coding is usually faster overall, but takes more instructions than the 'tl oop" method. Its feasibility arises from the
fact that the distribution of an input block into two output blocks
can change only when an item is added or deleted; in many files,
this may be only once in several hundred records. When it does happen, the memory transfer instructions must be adjusted to meet the
new conditions for input block reassembly, but they then remain constant until the next change occurs; this may be many blocks further
along in the file o
(3) The minimum time spent in memory transfers is easily determined.
Each transfer takes 0.051 l'Oo0036N milliseconds, where N is the
number of characters (and a multiple of 5). If X records are combined into Qne grouped record, the minimum total time for all trans~
fers necessary to reassemble one input block is X (00051 IO.0036N).
In the above example, where X = 5 and N = 200, this time is 3 775 ms.
Note that the minimum can never be achieved in practice, because at
some point address modification is required
Using a "loop·t to effect
the transfers can more than double this time.
0
0
One additional point of importance should be noted. At the time an output
block is ready to be ~itten, the IRe buffer contains an input record from
the previous read/write operationo This must be moved into memory before
the output can be wri tten-.-otherwise it will be destroyed-- and it can't
be put into the area already in use because that still contains a portion
of the previous input, the. last three records in this case.. Figure it any
way you like, the 705 either needs two input areas and one output, or two
output and one inputo One or the other choice must be made. One result
of this requirement is more instructions--and more computing time--either
to modify addresses or to transfer complete grouped records to a common
area.
The computing time necessary for the tape record reassembly, plus any
time required to make the tests to determine if master file items are
needed for processing, almost always can be done while the IRC is in a
read- write operation; consequently it often is not important that the reassembly time is quite variable o Total tape time, however, is capable of
1184.11
(3)
1184.11 (Continued)
relatively accurate determination and in many--if not most--file main~
tenance operations this is the critical factor. The minimum program
for the tape handling follows, with times in milliseconds.
SELECT
WRITE
TRANSFER ANY
READ
0604
2000
X
0.051
0.068
0.034
1000 0.051"L 0.009N
TRANSFER ANY
X
0.034
SELECT
0605
0.051
RWW
xxxx
0.034
SELECT
0602
0.051
WRITE
3000 0.051"L Oo009N
TRANSFER ANY
Z
0.034
Select applicable TRC
Write group mark; turns on indicators
Transfer if any error or special condition exists in the 705; this tests
accuracy of last TRC-to-Tape and
Tape-to-TRC transfers
If no error, transfer N"characters
from TRC to memory, beginning at
address 1000
To test accuracy of above transfer
Select input tape unit
Prepare to read-while-writing
Select output tape unit
Mode B3 write and read operation
Test accuracy of memory-to-TRC
transfer in Mode B3 operation
This is a standard program in IBM examples, and requires a total of
18. 459 ms of internal computer time, of which 90425 do not overlap
tape movement. (This time is based on the 1,000-character grouped
record). The actual transfer of the output grouped record (next to
last instruction) overlaps the start of the output tape, and the final
instruction also occurs during tape time. The total non-overlapping
time may be summarized as 0.425fO.009N; the total internal computer
time is 0.459 "L 0.018No
The total 705 time to write a block, read the next input, and do nonoverlapping computations may then be summarized, using a 1,000-character
tape record as the block size. 17"Loo067N for tape time plus 90425 computing time, a total of 93.425 ms for 1,OOO-character grouped record.
These are IBM times for a grouped record of this size, and they are
incorrecta As it frequently does, IBM has oversimplified in two respects:
(1) The minimum program shown above probably will not be used in normal'
practice; and (2) the transfer times themselves are incorrect.
For a discussion of why the program itself is probably oversimplified,
refer to the discussion on the limitations and implications of the "TRANSFER ANY'" instruction in section 5140 Consider now why the transfer times
are incorrect and note one statement: NONE OF THE TIMES IBM IS QUOTING
ON 727 TAPE TIMES OR 777-memory ARE ACCURATE.
0
First take tape time. The standard quotation for time is lO,Lo.067N, or
l7,Lo.067N ms. The correct time is 10fOo067(N"L4) or 17too067 (Nt4)o
The reason for the extra four characters is easy to seeo Tape read/write
operations are not completed until the record has been finished--and the
record consists of the data characters followed by three blank spaces and
the check character. Every tape record requires 0.268 ms more time to
read or write than published figures show. Let's see what actually
1184.11
(4)
1184.11
(Continued)
happens when reading a ta"pe record into the 'IRC buffer, for simplicity
taking only the last few characters of the tape record of 1,000 characters (the last then goes into IRC buffer position 999)0
Tape
Beginning of
Inter-Record G
IRC
Buffer
'£
~
Characters are transferred from tape to the IRC buffer through the record
mark in 999; recogni tion of the three "blanks it on tape causes the group
mark to be inserted into position 1,000 of the buffer and the recreated
check character into 1,001 after comparing with the character on tape.
Because the tape cannot miraculously ttjump over it the gap, it evidently
requires four character times to complete the checking operation. IBM
includes this in none of its tape time estimates--but'it is fundamental
in checking the validity of the transfexso
Similar reasoning will show that 2 character cycles of 0.009 ms each are
omitted from both IRC-to-memory and memory-to-IRC transfers. Thxough the
record mark, everything that happens is fine; this is the ttNn IBM uses ..
One additional chaxacter cycle must be added for the transfer of the
group mark to the IRC in memory-to-IRC transfers or for recognition of
it in TRC-to-memory transfers (remember that the group maxk does not get
moved in this case) .. One more cycle must be added for inserting the check
character in a memory-to-IRC transfers, or fox comparing the recreated one
against the original in a IRC-to-memory transfer.
The correct times for IBM tape and IRC-memory transfexs are based on
these formulas, not on those published by IBM.
Tape:
mc:
10 I 0.067 (N I 4) or 17 I 0.067 (N I 4) ms
0.009 (N I 2) ms (plus instruction execution time)
Using the corrected formulas, tape time for the transfer of 1,000-character records is 84.268 ms; total time in the example above is 94.711 ms,
9.443 ms not overlapping tape operations. Of the total time, 18.495 ms
are required for instructions and mC-memory transfers, leaving 75.216
available for useful computation during the tape read/write operation.
Total tape time, non-overlapping computations and the time available for
productive computations vary, of course, with the length of the tape
records in the file.
1184.12 Factors Pertaining to Univac II File Processing. In the handling of
fixed-length master items in Univac II, the same principles of block reassembly mentioned as possibilities in the 705 can be usedo There is one
important difference. Unlike the 705, the size of the master item has no
relationship to the output block, which is always 720 characters. Block
1184.12
1184.12 (Continued)
reassembly can be handled by Y-Z and V-W instruction pairs, changed
only as an addition or deletion to the file requires, or by movement
of individual items. This is also true of the 705; the choice of method
of reassembling input blocks into new outputs often is predicated in
large measure on what tests or operations have to be performed on the
i terns as the file passes through the EDPM. In general, ttloop't. methods
of reassembly would not be used unless the item size is an even submultiple of 60 words; if it is not, the straight-line block reassembly
may be preferable. Univac block time, of course, is a constant 51 ms,
and there are no non-overlapping instructions to be performed. As a
further note of interest, Univac does not demand the use of two input
and one output area, or vice versa, as does the 705; because rI and rO
are separate buffers, there is no necessity of unloading an input block
before an output can be written. Normally standby blocks are used to
permit input-output operations to continue during the occasional processes which may require more than tape time to complete. This does not
alter the fact that rI serves in many file maintenance processes as a
60-word extension of memory; the TRC buffer cannot be used for this purpose.
1184.13 Comparison of Univac II and IBM 705-777. Insofar as handling single in....;
puts of fixed tape block and fixed master item sizes is concerned, either
the 705 or Univac is relatively straightforward. Internally, such a file
is most ideally adapted to the high-speed internal memory transfer of the
705,and it appears that block reassembly can be done in about half-the
time of Univac; there will be some variation depending upon the number of
records per tape block in the 705, whereas this need have no effect on
Univac's internal time. In many file maintenance runs the speed of handling tape is more critical than internal speed, provided that all necessary internal work can be done within tape time. For both equipments, the
EDPM housekeeping necessary to test input blocks to determine if they require action, reassemble them into new outputs, and execute tape handling
instructions can be done in much less than tape time. This often involves
a fairly high percent of the total blocks in the file. Using Univac's
standard 5l-ms block time and the figures developed in 1184.11 for the 705
time with 1,000-character tape records, the comparative speeds of the two
equipments per 1,000 characters are:
Univac:
705
:
Thus Univac is
equipment (the
for the Univac
Univac does an
1,000 characters in 70.8 ms, or 14,300!second
1,000 characters in 94.7 ms, or lO,550/second
33% faster than the 705; using eight tape units on each
maximum possible with one TRC)~ the base monthly rental
is $22,140, for the 705, $21,4000 In terms of cost, then,
equivalent amount of work about 30% cheaper than the 705.
Because the 705 is hardly competitive with Univac II, either in time or
cost~ IBM is pushi.ng the ftdouble buffer" concept, using two or more TRes,
to obtain a better cost-of-perfor,mance figure. This will be considered
in detail in l186G
11840.2
118402
MAINTENANCE OF FILES WITH VARIABLE LENGTH MASTER ITEMS AND FIXED
LENGTH TAPE BLOCKS
1184.1 discussed in detail some of the aspects involved in the mainte~
nance of files with all items of the same length.. The more typical
business file has items of variable length, and the degree of variation
can be very large; files have been encountered in which the maximum size
of an item is 50 times greater than the minimum.. Because the EDPM has
no advance knowledge of how many characters or words are involved in any
one item, data must be included within each item which permits determining its length; some of the methods in use are discussed in 8110.
Insofar as this subsection is concerned, it should be noted that the
Univac treatment is practically identical with that of 118401; it
writes out in constant blocks of 720 characters which have no relationship to the item involved. Consequently, most of this discussion is devoted to the 705 ..
1184021 Factors Pertaining to 705(777 Processing.. In the fixed-item, fixed-tape
block method, the processing techniques with the 705 are quite similar
to Univac 9 s when the item size is a sub-multiple of 60--say 12 words.
The basic difference is that the block size with the 705 is adjusted to
an integral number of items, close to 1,022 total characters, and the
file then treated as of constant block size sizeg With different files,
the block size can vary, but it remains constant for any oneo The length
of the tape grouped records is then an integral and unchanging number of
items, and the end of an output record always comes in the same character
position, which is also the end of an iteme
Because constant-length tape records permit most efficient utilization of
the TRe in terms of tape handling, let us examine the implications and
feasibility of handling variable-length master items in a fixed-length
tape block. This means, of course, that a tape block will not, in general, end at the end of an item, but will terminate in the middle someplaceo Note that this is identical with the typical Univac situation
when variable-length items are being processedo Well~ it may be asked,
whatOs the difference in the 705? Why not handle it as we do in Univac?
And the difference is this: In the 705, internal memory transfers now
become much more complicated because of the way they are accomplished;
they are geared directly to a recordo Univac doesn 12t care how many or
what records are involved; it transfers a specific number of wordso
Before investigating what is now involved in 705 memory transfers, let us
note that block reassembly is still required
In fact, in the normal file
with variable-length items additions and deletions occur more frequently
tha'n in a fixed-i tern file.. Not only are we concerned with complete item
additions and deletions, but also with additions and deletions of the
variable sub-items within itemso In many files, these are of much more
frequent occurrence than changes in entire items" This is an inherent
characteristic of variable-length items, although the amount of file rejuggling naturally varies greatly in different types of applications ..
0
An understanding of the difficulties now involved in the internal memory
1184.21
( 1)
1184.21
(Continued)
is made easier by reference to a typical partially-filled output area
and the next input block, some of which is required to fill the output.
Both input and output blocks are to be of constant length, by definition,
and this is taken as 1,000 characters. The figures between record marks
indicate the number of characters in the record; as always, this figure
plus one (for the record mark) must be an integral multiple of five.
The output block is first.
tt- ~
<:)
a
~\n
~~
02-
1~9
~~
c.
259
~I
G
I B4
~I
H
2.
1..9
By previous operations, the output block has been filled with 570 characters, through position 3569, and 430 are needed from the input block to
complete it. This requires the first 30 characters of item F, in addition
to D2 (which is the rest of the D items) and all of Eo And because F is
105 characters long, including its record mark, it is obvious that the
high-speed memory transfer isn't going to stop after 30 characters, but
will transfer the next 75 into addresses 4000-4074, removing the group
mark in 4000 as it does so. And because the group mark is what stops
the writing operation, its elimination is undesirable o
Well, there are numerous methods of handling this undesirable situation.
In fact, just as in the discussion of reassembling the fixed item, fixed
block, there are so many possibilities that the development of a sample
method is somewhat futile; any specific file might be handled in a shorter
time. One way is to store the 30th character of item F, insert a record
mark in its place and do the high-speed transfer. Upon completion, put
the stored character in the last position of the output block. A second
method is to set one of the auxiliary storage units to the required number of characters in the last record, and move them one at a time 9 A
third is to establish the output area at a size large enough to hold
1,000 characters plus the longest item in the file; the entire item is transferred, overflowing beyond the output block size, after which the group
mark is restored to position 4000 to stop the writing operation. All--and
more methods--are possible, and all require computing time.
One statement, however, can be made with certainty, Straight-line coding
for the block reassembly can never be done; the loop method, with its
address modifications and tests for exit, must always be used. The ef~
fective speed of the high speed internal memory transfer then is reduced
because of the computing time required for "housekeeping" addresseso How
seriously this reduces the effective speed of internal transfers is a function of the average length of items, whether or not the items are moved to
a common "working storage" area for testing operations and productive work,
11840 21
( 2)
1184.21 (Continued)
the method of indicating the length of the master item, and similar
considerations. Internal memory transfers become, in effect, one
more highly important consideration to be taken into account with
the other complex factors associated with the determination of the
most efficient file format for a given processing.
There is one other facet of the reassembly process that deserves mention.
The maintenance of a fixed-length tape block in general means that part
of the final record in one block is carried over to the beginning of the
next. Furthermore, any record in the file can be so split, and usually
will over a period of time. Because its length must be an integral flUmberof five characters, including the record mark, in order to 'use the
high-speed internal memory transfer, it follows that the portion of the
record at the end of one record cannot be terminated by a record mark;
if it did, the combined halves of the record would have two record marks
and then not be an integral multiple of five. Needless to say, there are
ways out of this predicament, but--like so many other things--they take instructions and computer time. Referring to the illustrations on the previous page, what is required is a means of stopping the transfer of the
II partial record (the remainder is in the next input block) when 80 characters have been moved. Perhaps the most obvious way to stop the transfer
is to place in addresses 2000-2004 something like four Z9 S and record mark;
they would, of course, be transferred but could easily be written over in
the output area as the next inp\,lt block is reassembled. This costs only
one additional memory transfer cycle, but introduces another complication
which takes some computer time. Occasionally II will be a complete record
and have a record mark in 1999, which will stop the memory transfer. This
means additional programming--and time--to determine the exact nature of
the last partial record, which may turn out to be completeo
1184,,22 Factors Pertaining to Univac II Processing. Because Univac II's input/
output operations and high-speed internal memory transfers are both based
upon fixed numbers of words, either 60 or specified in instructions, the
fact that items happen to be of variable lengths causes no complications
in either type of instruction. Just as in the fixed-size item, it is perfectly possible and feasible to' adjust the memory transfer instructions
only when additions Qr deletions are made, and use them as straight-line
coding until the next change occurs. In other words, the internal memory
transfers of Univac are, or at least can be, independent of the information being trasnferredo
The use of standby blocks, block advancement in input and output areas,
and similar well-established techniques in Univac I installations can be
carried over bodily into Univac II practice; these methods likewise are
independent of the item size, whether it be fixed or va.riable ..
1184.23 ComRarison of the Univac II and IBM 705-7770 With either equipment, the
processing of' variable item length files requires programmed advancement,
based on data elements within the item, to step from one index number to
the next to determine if processing against a transaction is required;
this is also necessary if each item must be inspected for a possible action
dependent on something in the item (such as an expiration date for a magazine
subscription) and not upon a transaction ..
1184.23
1184.23 (Continued)
This requirement cannot be eliminated by repeating the high-order index
number at the first of the tape record, because, with block reassembly
involved, it would be necessary to redetermine this index number for
each new output block--and this would require "bootstrapping"' through
successive items in the input file. Consequently, both equipments require a loop-type program to advance through successive master items.
The 705, in addition to this, requires a loop to do block reassembly.
Univac does not.
The net result of this requirement is that internal memory transfers in
the 705 become, in general, more time-consuming than they are in Univac.
Although the internal transfer of the 705 is, nominally, about twice as
fast for an equivalent number of characters, the housekeeping involved
in address modifications and limit tests in the 705 may take more time
than the transfers themselves.
This leads to a significant conclusion: Unless record being transferred
of constant length, the so-called high-speed internal memory transfer
of the IBM 705 is, in general, no faster than that of Univac and in many
practical applications is slower.
This statement also highlights another
fact or general importance. It is fallacious to compare EDPMs on the basis
of the speed with which they execute comparable instructions. The important
factor is how long it takes to do a specific job.
21:§.
Insofar as ability to process the file is concerned, the comparative tape
speeds or 1184.13 apply here; there is no change in the minimum non-overlapping computer time 'needed by the 705. However, with variable-length
master items, the available computing time in the 705 is less than with
fixed-length items; in Univac II it is,or at least can be, the same.
In file operations in which the use of standby blocks is advantageous,
the usual condltion,Univac II has, in general, a clear-cut advantage.
It can always be handled with a Y6-Z6 pair, which requires 5.02 ms. to
complete; the 705 record-by-record transfer would necessitate a loop-type
program in which address modification and exit tests wQuldnormally exceed
the actual transfer time.
1184.3
1184.3
FILE MAINTENANCE WITH VARIABLE-LENGTH TAPE RECORDS
The third possibility of file maintenance with the IBM 705-777 combination is assembling variable-length master items into grouped records
consisting of the largest number of complete items which can be contained
in 1,022 character position, rather than attempting to construct fixedlength tape records. This possibility obviously dOes not exist with
Univac, which always writes and reads fixed blocks of 720 characters.
1184.31 Factors Pert inin
the ramifications
(in part involved in handling variable-record, fixed-tape block lengths
in the 705 were discussed. Some of the difficulties which arose were
centered around the fact that, in general an input block ended in the
middle of a record, and reassembly of the output block normally required
splitting up an input record into two pieces. Although the condition can
be handled, both instructions and computer time, not necessary in the Univac system, were required. The question naturally arises as to the possibility of recording information on tape in the form of full records,
letting the length of the tape block itself vary.
Initially, consider the internal ramifications now encountered. The first
to be noted is that an input grouped record consists of an unknown number
of records and characters, each record terminated by a record mark. There
is nothing distinctive about the last one; the group mark, it will be recalled, is not transferred from the TRe buffer into memory. Suppose the
input area is established as 1,022 characters, the maximum size a grouped
record could be. These 1,022 positions always have something in them, because a TRC buffer-to-memory transfer replaces only the characters actually transferred. Further assume that records can vary from 50 to 300
characters in length; this means a grouped record may be 1,022 characters
(in rare instances, of course), but also may be as short as 725 (in the
instance in which the next record is 300 characters). We can be sure,
without special provisions, that a specific item is the last one in a
grouped record only if that last item extends (in this case) into position 973, because then we know that even the shortest possible item cannot be included without overflowing the buffer capacity. We can also be
sure that if an item ends in position 722 or earlier that another item
follows it in the grouped record, because even the longest can be contained within the maximum of 1,022. But if any given item ends between
723 and 972, in this case, we cannot be certain whether or not another
follows; it depends cornpletelyoh the length of the item.
The immediately obvious way out is to say that a test on the index number
will indicate whether there is another item following, or just uhash"
from some previous, and longer, input record. A moment's reflection will
indicate that this may not be so simple as it appears at first sight. It
is barely possible--one of those 1 in 10,000 occurences-- that an index
number from some previous input actually occupies those character positions, and any detailed examination of the characters involved would give
a positive result. Adding a sequence test would be necessary to be certain, and of course analyzing several characters to see whether or not
they had the pattern of the index number would be quite a bit of programming. Transfer high or transfer low tests are certainly not much good;
data from some previous input could well give the same indication as an
1184.31
1184.31
(Continued)
Index number. Inserting a specific character code which never appears
any place else in the record is not, of itself, sufficient; it is again
possible that some previous index number actually does occupy the place
being tested. Testing of the specific character and the index number
(for sequence with the current input) would probably be satisfactory.
Just as effective, and possibly simpler overall, is a "grouped record
terminator" consisting of four Z's and a record .mark, for instance, following the record mark of the last information item in the grouped tape
record. The important thing is the test must be completely positive,
and it requires instructions and computer time to ascertain definitely
where a record from tape actually terminates.
The reassembly of the input into two outputs is subject to the same programming complexities of 1184.21; it would require a loop type operation,
testing each record to see whether or not it can be contained in the remaining portion of the 1,022-character maximum output block. Finally,
once the last item had been placed into the output area, it would be necessary to determine the last character position (i.e., the location of
the final record mark), and program-insert a group mark into the next
higher address; this is necessary to terminate the output writing.
Everything considered, the housekeeping necessary to use this method
appears to be fully as complex as with the fixed-block tape record in
lieu of special provisions to terminate the assembly of part of a
record into the output we substitute complexities of determining where
an input ends.
In terms of tape time, the assembly of information into variable-length
tape blocks introduces additional tape time, as compared with that which
would be requireD to handle constant blocks of the average size of a
given fileo This is discussed in 1183.2, and the remarks made there
are applicable in this case. How much additional tape time would be
required depends upon the variations in grouped record lengths which are
possible, and their actual distribution through the file. Typically, in
this case (very short tape records are not likely to occur, but ranges of
something like' 700 or 800 minimum to 1,022 maximum), the percentage increase will range from 5-15% over that required to handle constant blocks
of the average size contained in a specific file.
For many typical files, it appears that the '''housekeeping't, involved is
about the same for fixed-length and variable-length tape records; both
'r'equire considerable programming and computin9 time. Because the latter
method always increases tape'tli:me·(see 1183.2), it would be expected that
the fixed-block tape record would be preferred. Although this appears to
be a perfectly straightforward and logical statement, it is remarkable
that no example known to have been published by IBM adopts this approach.
As a further note~ it may well be that the difficulties of handling
variable-length records may be the reason that IBM advocates--and most
of its examples are based upon--fixed-length records.
1184.32 Factors Pertaining to UniyacProcessing.
in this manner.
None.
Univac cannot operate
1184•. 33
1184-.33
Comparison of Univac II and IBM 705-777. There appears to be little
to be added to the comments in 1184.23 on, variable-item, fixed tape
record fileso Under either approach, the characteristics of the
internal memory transfer and writing operations on the 705 pose difficulties and require programming not necessary with Univac. Of the two
possible approaches with the 705, there seems to be little to gain, and
some tape time to lose, with variable-length tape records.
1185
1185
FILE MAINTENANCE USING "MICROSECOND INTERROOATION" (IRe MODE D)
The processing of many business files involves the searching of a large
volume of master items to find the relatively few which are subject
to action because of some form of transaction input--and order for a
stock item, a change of address for a magazine subscriber, etc. Many
blocks of information may be passed over before the next master item
is found. TRe Mode D, called "microsecond interrogation" by IBM, is
a form of operation included to speed up passing over the unneeded
information in a master file; refer to 1182.4 for the method of operation of the TRe in this mode.
Because Univac is not capable of operating in this fashion, this section
will discuss in the detail necessary the implications of this mode in a
705-777 installation.
1185.1 File Maintenance Facts Inherent in ''"Microsecond Interrogation. It- The basic
fact to be noted, with respect to tape files and operating speeds, is that
only those grouped records which contatn a master item to be processed
are transferred from the TRe buffer into the 705 memory. Those grouped
records which do not contain such an item are written directly onto the
output tape without going through the memory.
An immediate consequence of this method of passing over the master file
is that reassembly of the input blocks into new outputs is impossible.
Put another way, all additions and deletions to the file, whether of
complete records or portions of them, must be done in the grouped record
in which the master item occurs
This is fundamental, and is applicable
to master files of both fixed and variable item length.
0
In practice, what this means is that, as the file goes through several
cycles of the maintenance operation, the records becQmemoreand more
variable in length, and tape time increases, as discussed in 1183.2.
For example, suppose the file starts out with all items grouped into 1,000character tape records. If an item is deleted from one of these records,
the remainder may be, say, 900 characters. If an item is to be added,
it is inserted where it belongs~ and, say, 100 characters from the end of
the record are written out as a'new and separate tape record. As time goes
on, the variability of records on tape in the file increases and so does
tape time, until ultimately the eniiremaster file is rewritten to restore
the tape records to approximately equal size. The increase in tape time
is rapid; for example, if a file of 10,000 grouped records of about 1,000
characters each has added to it 100 items of 100 characters each, the increase in the number of characters in the tile is only 0.1%, but the increase in tape time is l%--ten times as great. And a similar condition
holds for deletions; a drop in the file size of 1% may change tape time
by less than 0.1%. The effects of combined additions and deletions,
which is the normal expectancy, is even mOre spectacular, The net file
size can remain unchanged, but the tot.al tape time can increase by 15% to
25% or more, within a very few cycles of processing. How rapidly the tape
time increases is rather evidently a function of the percentage of additions
and deletions which occur during the successive cycles of processing.
1185.1
(Continued)
1185.1
The above discussion permits the establishment of certain criteria
which a file should meet to be susceptible of maintenance in the
"microsecond interrogation" method:
(1) The percentage of additions and deletions of information in the
file should be extremely low per maintenance cycle--no exact
criteria can be established, but well under 1% seems desirable.
(2) Most likely, the file should consist of fixed-length master items
in the sense that necessary information in each item is practically
a constant number of characters. Any "paddingta: necessary to make
all items constant simply adds to tape time. The reason for the
fixed length item, rather than those which are inherently variable,
is that additions and deletions are usually much more likely in a
variable-length record than in a fixed one; in the latter, entire
items must be added or deleted. In the former, changes within items
must also be considered.
(3) As a corollary of the above, the processing involved in the file
maintenance should only change data elements already in the master
items, not add or delete information.
(4) All processing to be done must be initiated by transaction inputs,
rather than by information within the master item. The only exception would be those cases where criteria for possible action
based on something in the master item could be included in the 8,
16, 32 or 64 characters moved into memory. This is self-evident,
since nothing can be done on data in the TRC buffer ..
These are basic requirements of the file which must (or at least should)
be met before microsecond interrogation can be considered as an economicalor practical method of file maintenance. One additional factor-volume of activity--must also be met; this is discussed in the next
subsection.
Parenthetically but importantly, it should be noted that very few business files meet all four of the-:.criteria for microsecond interrogation
which are listed above.
1185 .. 2
Level of Activity in a File Maintained-by ''Microsecond Interrogation.'"
IBM is advocating ~crosecond interrogation" file maintenance for low
activity file.§.., but so far as known has published no definitions of what
it means by ttlow't" or at what point some other procedure may be more
suitable. These criteria will be developed in this section and an explicitmathematical formula developed to permit ready determination of
when a file can be processed more economically by this method than by
some other.
To begin, consider how a 705 must maintain a file through successive
cycles in this mode of TRC operation. A grouped record from an input
tape is read into the TRC buffer at the sanm time the previous record
is written out. The first 8 (16, 32, 64) characters are transferred to
memory and tested against the next transaction input item to find out if
the corresponding .rnaster is in this grouped record. If not, the buffer
1185 •.2
1185.2
(1)
( Continued)
is written out and simultaneously the next input block read in. There
is only one mode of operation to do this simultaneous read-write--A3
(see 1182.13).
If a given grouped record in the buffer contains a master item required
for processing, the entire group is transferred into memory, the computation performed and the updated grouped record transferred back into
the TRC buffer. Then the combined write-read begins. Puring the entire
TRC-to--memory, computing, andmemory-to-TRC operations, no tape movement ~
can occur, regardless of how long or how short the computation. Obviously,
nothing can be written out, because the last previous input record has
already been so handled.. Likewise, the next input cannot be read into the
buffer. Why? Because it would have to be moved into memory in order to
write out the block requiring computation-- and moving blocks into memory
is precisely what umicrosecond interrogation" seeks to avoid. Therefore
whenever a grouped record requires 'Qrocessing, tape movement stops.
We can summarize the information necessary to establish a formula to permit determination of the economy of this method of file maintenance. The
time saved by "microsecond interrogation" is the time necessary to transfer a complete grouped record from the TRC to memory and back again for
every input block; the time lost is that required for computation, during
which period no tape movement can occur. This is the general statement;
it will be defined and developed more explicitly in the next few paragraphs.
First of all, the specific formulation desired is one that will enable a
determination of which is faster, "microsecond interrogation'" methods or
one which brings all of every tape record into memory; the latter method
is that discussed in section 1184, and involves TRC Mode B3. In both
methods, certain instructions do not overlap tape movement, and it is
well to see what differences, if any, exist. First, consider a sequence
of instructions to handle a file by "microsecond interrogation".. It begins
at the point where the input grouped record is in the TRC buffer.
SELECT
0609
WRITE
nun
TRANSFER ANY
X
1000 0 .. 051,LOo009N
READ
Set up TRC Mode D
Write group mark (to turn on indicators)
Transfer if any error exists
Transfer N characters into memory
N 8, 16, 32 or 64
y
COMPARE
0.034,L0.017M Compare index numbers, M is length
or index
TRANSFER HIGH Z
00034
Detail is greater than Master
TRANSFER EQUAL A
0,,034
Detail is equal to Master
SELECT
0602
0 .. 051
Select input tape unit
RWW
xxxx
0.034
Prepare to read-while-writing
SELECT
0601
0.051
Select output tape unit
CONTROL (WST)0007 0.051
Write-read in Mode A3
Transfer to another program
=
This is the minimum program to accomplish the work which cannot overlap
tape time. It is noted that the time is variable in two respects: (1)
1185.2
(1)
1185.2
(Continued)
The number of characters (8, 16, 32 or 64) transferred from the TRC
buffer into memory, and (2) the number of characters in the index
being tested, which can be any number up to and including whatever
was moved in from the TRCo However, the minimum and maximum times,
together with three intermediate ttaverage" times, based upon index
numbers of 8,12 and 24 characters, are summarized below:
~
Minimum
Average
Average
Average
Maximum
(I-character index)
I (8-char. index)
II (12-char. index)
III (24-char. index)
(64-char. index)
-1L
...M...
8
8
1
8
12
24
64
16
32
64
Time
~ms)
0 .. 582
00701
0 0841
0 919
2.157
0
The range of 0.7-0.9 ms can be considered most typical .. Parenthetically,
the program on the previous page is taken exactly as presented by IBM in
its manual on the Type 777, and has a rather obvious omission, which required one instruction and 0.034 ms more to execute: As presented by them,
the accuracy of the transfer from TRC-to-memory of the control word is not
tested; the first ''READ'' instruction should be followed by a ttTRANSFER ANY"
to do this. So even this minimum program is not sufficient.
The development of a formula to express the sum of tape time and non-overlapping computation is quite simple, and is accurate provided all tape
records are of identical lengths., Let ttL'" be the number of characters in
the tape grouped record, "N" be 8, 16, 32 or 64, choosing it to be the
smallest that will contain ftM tt , the number of characters in the index
number. Recalling that tape time in Mode A3 is 17 I 0.067L, and adding up
the constant and unchanging times from the program above (including IBMvs
omitted 0.034 ms), the following formula gives total tape passing time.
17 .. 527
I
0.067L
I
Oo009N
I
OoOl7M milliseconds.
Typically, this will be 84.7-84.9 ms~ The entire file, of course, is not
processed at this rate; this covers only the time to pass over tape blocks
not required for processing. To total tape time must be added the amount
of non-overlapping computationo
This is, of course strictly a function of the processing required and how
often it occurs. Before discussing it, a few remarks on "low" activity
in an EDPM are in order. Although IBM advocates "microsecond interrogation" for "low" activity files without further defining the word, in many
business applications, a file maintenance process in which input transactions run from li-3}b of the number of items in the master file is low.
With a transaction density this miniscule, "microsecond interrogation tt
sounds attractive. Then suddently it is realized that several master items
are grouped together in one tape record--it may be only a couple, but may
be eight or ten. And it further becomes apparent that with a transaction
density of, say, 2)6 and six items per block, something short of l2?b of the
blocks Ydll ~ involved. (It won't be quite 12, of course, because some
blocks will have two or three transactions to be posted into them).
1185.2
(Continued)
For the further development of the formula for the total processing time,
it is necessary to assign symbols for other quantities involved; the following are all involved:
L = Length (or average length) of the grouped tape records.
M = Number of characters in the index number
N 8, 16, 32 or 64, whichever is the smallest that will contain M.
T = Total number of Transaction items.
X = Number of items in the master file.
Q = Average number of master file items per tape record.
P T/X - Transactions as a percentage of the master file.
K = Average processing time per transaction item processed.
=
=
The additional time involved is that necessary to move T input records
into the memory from the TRC buffer, plus Computing time. This assumes
that each transaction will affect a different tape record; although some
tape records will have two or three transactions to be posted, the total
number will be quite close to the number of transaction items. This is
particularly true when transactions, as a percentage of the master file,
are very low; if they number 2% of the size of the main file and, on the
average, affect an item every eight or ten tape blocks, there usually will
be very few blocks having two or more transactions. The formula for the
total processing time then is:
(17.5271 0.067L/ Oo009N
1 0.017M)
X/Q
1 T(O.051/0.009LI K)
If no·t known, T, the number of transaction items, is equivalent to PQ, and
both of these are usually known. Actually, knowing P implies that T is
known, but people will usually give an estimated percentage of transactions, even though they may not be willing to state a definite average
number.
The expression above can be adapted to any given file maintenance process.
The amount of actual processing timet K, usually will not be known, but
often an EDPM technical consultant can estimate it within 15-20% by a not
too time-consuming analysis of the job, and an estimate accurate within this
order of magnitude is probably more than is justified; the prospective customer will change his mind many times before the job actually goes onto a
computer ..
Although use of the formula developed above will always give the time required to maintain a file by ttmicrosecond interrogation,tt it may be more
useful to obtain a general indication of the co'ndi tions under which it is
potentially applicable. With the 705-777 combination, the alternative
method is to read every record into memory, which of course takes more
time than merely reading a control word of 8, 16, 32 or 64 characters.
But now the computing time, K, in many instances completely overlaps the
tape read-write operations. If the percentage of transactions is high
enough, the lattex course may be faster, overall, than "microsecond
interrogationo" The three variable factors to be considered are length
of individual records, percentage of transactions, and amount of computing time (average) per transaction. In the next section, 118503, is developed a graph relating these three variables for various combinations and
values and explaining the use of the graph.
1185.3 Determination or Processing Time Savings Using "Microsecond Interrigation"
90%
80%
70%
60%
50%
\
\
\
\
40%
30%
20%
,,
,
" "-
10%
"-
"
-
---...
1'4:::
o
10
_
N:. 2.,.00
--- -------- ....
-...-
100---------------------
20
30
40
50
60
Amount of Comput.ation per Transaction (ms)
The chart above gives ;ttbreak-even tt curves which permit a fairly close
approximation of whether tfmicrosecond interrogation": is a faster means
of file maintenance, using the 705-777, than entering all master file
items into memory. Five curves are shown, for tape records (or grouped
records) of 100,200,500,720 a'nd 1,000 characters. For a given amount
of computation per transaction, "microsecond interrogation" will be
faster if the percentage of transactions is below the curve; If the
percentage is above the curve, it will be slower than reading everything
through memory. Tabular values on which this chart is based are included
on the next pageo
For both types of processing, Ims has been allowed for essential machine
housekeeping and tape movement instructions; in practice, this is seldom
enough if the file is read into memory. The basic effect is that the
amount of computing time shown must include, in this case, that which is
required for block reassembly and similar types of operations--anything
that will cause the 1 ms allowance to be exceeded. It will be noted that
70
1185.3 (Continued)
1185.3
(.2)
each curve becomes a straight line when computing time reaches a certain
point; this point occurs when the available computing time for a tape
record of a specific size equals the amount of necessary computation. For
example, with 100 character records, the available computing time during
the tape handling of one pair of records is 21.9 ms; if the actual computation requires more than that, the 705 is computer limited in both cases.
It is important to note that the percentage points are based on blocks,
not on transactions. For example, with 1,000-character tape records and
30 ms of computing per transaction, the break-even point between the two
methods is 25.1% of the tape records requiring processing. If there are
two master items per tape record, this corresponds to a transaction rate
of 1105%; if ten items per tape record, to a rate of 2.3%. The percentage
figure used in this chart are equivalent to the transaction rate multiplied by the average number of master items per tape record.
Tabular value for this chart are given below: ItKIt is the computing time
in ms and "%U means percentage of tape grouped records requiring processing.
1,000
720
100
200
500
_K_
_K_
~
.JL --2L
.JL
~
1.JL
~
43.0
5
56.6
5
64.4
5
27.0
16.0
5
5
31.0
10
38.8
10
47.5
8.7
10
10
10
1506
20
24.2
20.
31.2
18.5
20
6.0
15
11.0
15
30
17.6
30
13.2
30
23.1
20
20
8.5
4.5
10.4
40
13.8
40
18.4
40
r; 21 .. 9 4.3
25
6.9
50
15.3
50
11.4
~26.8
6.7 ~~1.5 : 9 .. 9
60
13.1
~52.3 10.9
~66
12.0
In general, the chart and table may be considered as giving the maximum
tape record processing percentage and in most practical applications the
actual break-even point will be somewhat lower than that indicated. One
obvious reason is that a master file processed by "microsecond interrogation" may have duplicated index numbers in the control field, thus adding
to the size of tape records and tape time when compared with the method
of reading everything through memory. However, both are useful as a
simple means of estimating approximate break-even points for the two
types of file maintenance processing.
1185.31 Comparison With File Maintenance Using Type 754. IBM has been promoting
"microsecond interrogation" for low-activity files, as one means of speeding up overall processing. They have been a good bit more negligent in
pointing out that use of a 754 is still faster in this type of operation.
As a matter of fact, this statement can be made categorically. The use
of a single TRe in "microsecond interrogation" file maintenance is never
as fast as a single Type 754 (unbuffered); the latter is ALWAYS about 6ms
per block fa~. The reason is that the 754 does not impose an automatic
tape stop between blocks, but can cover the gap in 10 IDS plus essential
1185.31
1185.31 (Continued)
non-overlapping computation, which will average I-It ms; this is about
II-lIt ms per tape record compared with 17 ms using the 777. In both
cases, all processing is outside tape time, and it is rather evident
that the TRC, in a "microsecond interrogation" mode, can never catch
up with the unbuffered 754.
An obvious extension is the fact that using the TRC in the mode of reading all items through memory results in break-even curves, when compared
with the 754, which have lower percentages than the comparison of the two
modes of TRC use. The reason is rather evident: For every tape block,
the 754 "gains" about 5!-6 ms over the TRC, and the accumulated gains
can be used to offset computations when these are required. It also
gains additional milliseconds which the TRC buffer requires to unload
its contents into the 705 memory; with 1,OOO-character tape records,
this is another 9 ms. If a file is established with tape records of
this size and the amount of computation per block is under about 16 ms,
the 754 will process the file faster than the TRC even if 100% of the
blocks are involved. It does not appear that IBM has ever presented its
prospects with any comparisons of the relative merits of the unbuffered
Type 754 and the buffered TRC. From these few remarks, it is rather
evident that the mere fact that a buffer is available does not automatically mean decreased processing time; as a matter of note, many
business file updatings can be done faster through use of a 754 than
with a TRe.
These comparisons are applicable only when discussing a 705 with ~ TRe
or one 754; as soon as two or more TRes enter the picture, the 754 loses
much of its advantage. This is explored more fully in 1186.
1185.4
1185.4 PossIble Duplication of Information in an Interrogated File. The fact
that ttmicrosecond interrogation" permits testing c{)ntrol or index numbers
only from the front end of a grouped record on tape means that, in abnost
all normal applications, a duplication of the high index number from the
group at the first of the record. This is rather evident, because in testing a transaction input item against the master file, it is necessary to
know which tape record it is in, and this cannot be done simply by looking at the first of a group of items in one tape block. If the high-order
index number is duplicated, total tape time is increased by a percentage
directly proportional to the length of the duplicated index number as a
percentage of the remaining characters in the tape grouped record.
In some instances, this proviso may be unnecessary; for example, if each
grouped record consists of the same number of items as all others, and
the index number is short enough, it may be possible to put all of them
in the first 64 character positions, or less, and not repeat them within
the block. This is not quite so desirable if different grouped records
have varying numbers of items, because now it is necessary to determine
which index is the last in the tape block. In some files, it may be
feasible to arrange items within the block in the sequence reversed
from the main block-to-block sequence; ie., to arrange them in descending
sequence in the block, although the file itself is handled in ascending
sequence. This places the high index number as the first item in the
block, and removes the necessity for duplication. Other possibilities may
exist in files of a specific nature, but as a general rule it may be stated
that some redundancy--adding to tape time--will exist in files processed by
"microsecond interrogation".
The second source of duplication of information which may arise is in the
area of what may be called, to relate it to conventional punch card practices, "header data." For example, in a name and address file arranged
georgraphically as primary sequence, the most efficient magnetic tape
format places the city and state only once at the beginning of the items
pertaining to it; it is not necessary to repeat the city and state, or a
coded abbreviation, into every master item. This may not be possible in
a file processed by "microsecond interrogation, .t without the inclusion
of special progranuning (for recognizing a change in city) which may add
to tape time by placing a short tape record whenever the change occurs,
or by additional testing of the control field, which does not overlap
type movement during the testing of the tape record to determine if it
is required for processing.
As a general rule, any proposal to use "'microsecond interrogation" with
the 705-777 implies some redundant tape information, or additional nonoverlapping programming, either or both of which add to the total time
necessary to process a given master file. The degree of such duplication
may vary widely, but it is pertinent to point out to potential users the
fact that it almost always will exist.
1186
1186
USE OF TWO OR MORE TYPE 777 TRes IN THE IBM 705
The previous few sections have discussed .in some detail the basic principles of operation of the Type 777 TRe, and factors pertaining to its
use in file maintenance operations using both '·microsecond interrogation"
and the method of reading all information into memory. Using either procedure, the 705 is inferior to Univac II, a major re.ason being that Univac
has an effective tape speed at least 30% faster than the best that can be
obtained in a 705 installation. With one TRe, the 705 is simply not competitive with Univac lIon a cost-of-performance basis.
For the past several months, IBM has been strongly advicating the use Of
two (and in some instances, three) TRes to permit multi-programming of EDPM
operations.. Actually, there are several reasons for this sort of proposal,
and IBM is avoiding mention. of some of them.
(1) With two TRes, it is possible to split a file into two halves, and
process the two halves in parallel, thereby cutting file maintenance
in half. This is, of course, the point IBM has been stressing.
(2) IBM has advocated multi-plexing of two different jobs in exactly the
same manner.
(3) With one Type 777, the 705 can handle only eight tape units; with one
unbuffered Type 754, only ten. With the larger memories of presentday
EDPMs, it is often desirable for an installation to have more than this
number. This means at least two tape control units for the 705. IBM
has not stressed this point.
( 4) The cost of an installation processing a file in sections is greater
than that of handling it as one file. This fact is not stressed except by the implication that fttwice as much work can be done;'" there
is no attempt made to demonstrate this point.
This section will be devoted to an analysis of the two-TRe, or fldoublebuffered" operation of the 705. The development draws heavily on the
discussions in 1183-1185, which should be understood before attempting
to read this section.
1186.1
(1)
1186.1
Implications of "Double Buffer" Operation Using Two TRCs. In the
double-buffer approach, the various methods of handling a file maintenance process with one TRC.can be applied to two, doing the job in
parallel, or the two TRCs may be operating with a different technique-"microsecond interrogation" in one and reading everything into memory
with the other. All of the comments made on the various types of
operation with one TRC apply to the use of two, and additional factors
must now also be taken into consideration.
Consider, initially, the objective of the double-buffered approach. To
permit overlapping the movement of two sets of tapes. Timewise, this
may be depicted graphically, assuming that the tape records from both
sets of tapes are read into memory from the TRC Buffers.
AV!lJZJ
------------------------------------------4
E
tzlc
WIlAD
f7/I //!ImJIZ7A
E
F
During time A, the 705 is occupied in testing an input tape record, dumping from the TRC buffer into memory and loading the output record from
memory into the TRC; in mode B3, tape advance begins simult:"lneous with
loading the buffer. During time B one record is written on an output
tape and one read in to the first TRC buffer. As soon as time A is
completed, the 705 checks to see if the second TRC has finished its
previous operation; this is time C. If it has, it enters into time D,
durihg which it-unloads the second TRC buffer into memory, tests it for
action and reloads the buffer, Simultaneously starting the output tape into
motion. The second TRC writes one record from buffer and reads another in
during time E. As soon as time D is completed 1 the 705 is free to do useful computation until the first (or second) TRC has completed its writeread operation. Thus the two tapes from two different TRC move in parallel, with, of course, a slight time lag between them. This is the basic
theory of the ope:ration, and IBMs claims are that the 705 is so fast internally that times A and D are so short that sufficient computing time
(E) is normally available to permit processing to overlap the dual tape
read-write operations. It now remains to examine the implications of
this method of operation.
First, there is the matter of-control--that is, setting up a routine to
switch back and forth between the TRes. This can be quite short, because
it is possible to test the status of the TRC (1182.7). The method is to
select one tRC and execute a "Transfer Ready·t: instructionJ if the IRC
has completed a read-write operation, transfer to the processing routine.
If it hasn't, test the other in the same fashion; if it is not ready, go
back to the first. Thus, when no computation is being performed, the 705
1186.1
1186.1
(Continued)
continues cycling through this control loop until one or the other of the
IRes has completed a previously-initiated read-write operation. In the
time-sequence chart above, this loop occupies time C and, using IBM-suggested programs, it lasts about 0.2 ms.
1186.2
Factors to be Considered in A Double-Buffer Process. A double-bOffer
operation on the 705 is, fundamentally, nothing more than combining two
single-mC processes into one machine run. So far as the work being done
through each me is concerned, it is subject to all of the considerations
discussed in 1184 and 1185, depending upon its mode of operation. In addition, the joint use of the two mcs introduces new factors caused by theIr
interaction. These axe superimposed upon the structure already developed,
and this section is devoted to an analysis of these new factors and their
effect on the machine operation.
1186.21 mc Control Sequences. A ,single mc program ~an be thought of as being
divided into two 10gical.,\;phases: (1) The tape handling or input-output
routine, which may als6include master item testing against a transaction,
and tape block reassembly, if applicable; and (2) the processing routine,
which performs the computations requtred when a master item is to be processed. In a double-buffer program, these two phases exist for each IRe.
One more phase also is added: A "'control sequence" to provide the linkage
between the two mcs. These program phases may be depicted graphically:
Processing
Routine A
Control
Sequence
Processing
Routine B
These deliberately are unconnected. Not only are different linkages possible, but what is done in each of the phases is subject to a wide amount
of variation. As an example, the two input-output sequences may be so
linked that there really is no "control sequence" as such.
1186.22 Testing of Master File Items. The tape records from each IRC must be
examined to determine if processing is required; this may be limited to
matching against an input transaction item, or may involve tests on data
in every master file item in ,the file. There may be a separate test sequence for each mc, or acornmon sequence for both. In the latter case,
a "control sequence tt is necessary, in some form, to return the program to
the proper place when testing is completed.
1186.23 Input-Output Areas.- Depending upon the mode of processing adopted (i.e.,
"microsecond interrogation t ' or reading everything through memory), dupliOriginal 1/31/57
1186.23
1186.23
(Continued)
cate input and output areas in memory may be necessary. The amount of
duplication could be fairly small--in "microsecond interrogation," for
instance--or run into several thousand characters. It !:also should be
noted that such duplication is not limited to the master file, in the case
where it is processed in two sections, but may carryover into several
subsidiary input and output files.
The import of this duplication of memory areas :ls:,~t on¢:~· @bvious: Whatever duplication is necessary effectively serves to reduce'the size of
memory below that which otherwise is available. A direct effect is that
less memory is available for instructions.
1186.24
Tape Units. If one long file maintenance run on a master file is performed bysplitting it between two IRCs, duplication in tape units must
exist. Instead of four for the master file (two input and two output,
both sets "ping-ponged), there are eight--four for each IRC. Instead of
one transaction input tape, there are two. Other input and output tapes
may also be duplicatedo The net effect is either to increase the total
number of tape units in an installation, or to reduce the number available
for different purposes in a double-buffered machine run.
1186.25 Multiplexing of Processing. The use of double-buffer processing may consist of splitting one file into two sections and processing both sections
"in parallel, doing exactly the same type of work on both halves. Or it
may consist of doing two completely different jobs in one machine run,
each IRC controlling the tape units for each job, and the two separate
programs tt.time-sharing tt the available EDPM computing time.
1186.26 Machine Runs Before and After a Double-Buffer Process. A double-buffer
machine run may impose specific requirements on preceding or following
runs, or may require creation of runs not otherwise necessary. For example, a tape of daily transactions must not only be sorted into sequence,
but must be split into two parts, on separate tape reels, corresponding
to the division of the master file; controlling the split may add to sorting time. Similarly, several minor outputs which could be written on
separate tape units in a single-buffer type operation may--because of
duplicate tape unit requirements, for handling the master file--need to
be consolidated on one output tape, and separate for printing (as an example) in a completely new machine run. All EDPM spent in this sort of
extra wo'rkis, of course, properly chargeable as a cost of a doublebuffer operation. '
Original 1/31/57
1186~3
(1)
1186.3
"Microsecond Interrogation" Processing With Two TRCso In this mode of
operation (see 1185), the internal 705 time necessary to examine one tape
record to determine whether or not it requires processing and to execute
the instructions for moving the tapes take relatively little time with
average-length index numbers--something between Oa8 - laO ms is enough,
using IBM suggested programming. It then appears to be quite attractive
to use two IRCs in this mode of operation, because only about 2.0 ms are
required for internal tape handling and testing computations out of the
85 ms available in reading and writing tape blocks of 1,000 characters.
Such use of two--or even three--TRCs offers an attractive means of increasing the effective number of characters-per-second handled by the
705. Let us now examine more closely this superficial statement.
First of all, the discussion of 1185 pointed out that few business files
have characteristics making the "microsecond interrogation t• approach
particularly appropriate. Even if the tape records are periodically regrouped into ones of constant length, additions and deletions soon cause
them to vary, and tape time per thousand characters goes up. These facts
should be understood clearly.
Now the splitting of such a file between two .IRCs does not quite double
~he transfer rate from tape units.
One new factor at a minimum is added:
The possibility that one IRC becomes ready and must wait, while an item
in the other is undergoing processing. Note that is is logically highly
impracticable to attempt to interrupt computations simply to keep the
first IRC going. There are two reasons for this; (1) The length of tape
blocks is variable and there is no foreknowledge of how long it will take
to read any specific 1'next record, tt and (2) because one set of tapes is
stopped during computation, the two sets do not move synchronously and even
if all tape records were of a known and constant length, the relative
1tstarting pointst' of two corresponding blocks changes every time a computation occurs. A time illustration similar to that of 1186.1, may facilitate understanding this phenomenon:
UA
~_________W_r_i_t_e__a_n_d__
R_e_a_d_IR
__C
__A
__T_a~p_e_s____________________~)
Available for Computation and Control~e$!Jence
contro[l
~I____________
cD I
DB
~
B Write and Read IRC B Tapes
D
~ ~I--------------------------------------------~}
~I-----------This represents the initial starting condition of the two halves of the
master file, with records of identical lengths being handled. The tape
handling sequences for the two IRCs are designated. by A and B. With no
computation required, the time available for com.puting is spent in cycling
through the control sequence until (in this case), IRC A is ready with
its next record. Suppose it requires some action; this is the time sequence now:
Process TRC
A Item
Write and Read mc A Tapes
OA
'1~'--~--------------------~--------------~) rl----------
)
cO I Computing Time
)
and
I IControl
DB Jiead and Write DB
I
>
Sequence;
I .I. ._____{-.
IRC B Tapes
~I---------------------------Original 1/31/57
1186.3
(2)
1186.3
(Continued)
Tapes from IRC A, of course, are stopped during the computation period;
this is inevitable. But, likewise IRC B is held up, because it is ready
with its next tape block before the 705 is ready for it. Now this tape
stoppage isn't going to happen every time an item is processed. In the
above case, for instance, an item from IRC B could be handled without delaying the tapes from IRC A unless the amount of computation were rather
long or the next tape record from IRC A were rather short. However, there
is going to be some variation in record lengths, and even more variation
in the amount of computing time necessary to handle each item which requires processing; the special sub-routines necessary to handle each
type of transaction make that certain. It is thus quite evident that the
movement of the two sets of tapes is not going to be synchronous and that
some computations are going to stop both IRCs. In other words, there will
be some interference between the TRCso Less obvious, but equally true,
the effect of this interference on total processing time is going to be
difficult to determine; it depends not only upon the amount of computing,
which usually varies from item to item, but also upon the variable lengths
of tape records. Some computations will be performed with no loss of time
on the other IRC; others will stop if for almost the full processing time,'
as inihe example above. An actual estimate of the time loss is not too
important, because if the process is a low activity file maintenance of
the type suggested by IBM as sui table for umicrosecond interrogat'i;Qf1\t'~:I,·
the interference time is only a small percentage of the tape time.
It may be illustrative to consider a hypothetical example of a 100,000item master file, grouped five per tape record averaging 750 characters
each. Assume 5,000 transaction (a 5% transaction density) are to be processed (involving 25% of the grouped tape records), at 20 ms processing
each. Using an 8-dt'gtt index number, these times can be established:
Read 10,000'grouped records on one IRC at 67.52 ms per second
(this is half the master file).. .. ..
Non-overlapping tape-handling instructions and control
field interrogations •• e ..
0
Processing time, 2500 items at 20 ms each
Non-overlapping TRC-to-memory transfers, 2500 @ 6075 ms
each.. • • .. .. .. • • .. .. .. ..
10% (assumed) variable record length allowanceo ..
Interference factor with other TRC: Assume 33% interference
probability on 2500 records, average delay of 1604 ms (onehalf of processing time plus TRC-to-memory and memory-toIRC transfers)" ..
0 ..
0
0
"
"
0
..
0
It
..
....
It
....
0
......
Q
..
"
....
"
It
0
•
"
..
..
..
0
..
0
0
......
Total time
0
....
It
..
..
•
.
....
"
675.2 sec.
7.0
50.0
16.9
67.3
13.7
830,,1
The processing of the other half of the master file, from the second IRC,
overlaps this time.. Thus two grouped records are processed every 83 ms,
or one in 41.5 ms; this is an average rate of 1,000 characters every
55.3 ms, as compared with a theoretical maximum of about 42.5 ms~ With
a single IRC, the time per 1,000 characters is not quite d~uble 55.3 ms;
the saving is the interference factqr which amounts to only 0,,13 ms per
Original 1/31/57
1186.3
(3)
(Continued)
tape recordo It appears, then,that the use of two mcs nearly doubles
the EDPM output at an increa~in equipment rental of 26-34% (depending
upon the number of tape units associated with each mc)o
Realistically, the percentage improvement should be based upon the performance possible with one unbuffered Type 754 Tape Control Unit, which
is always faster than a single mc operating in the "microsecond interrogation·' mode. Th.e file postulated above could be processed in about
70.7 ms average per tape grouped record; now the 41.5 ms possible with
two IRCs represents about a 71% increase in production at a 30~40% increase in equipment rental. Additional factors discussed in 1187 will
show that even this percentage increase is never attainable in practice;
there are many EDPM operations which will halt the entire operation regardless of how many IRCs are involved ..
Most 705 installations can be expected to have two mcs; although one
754 might provide enough tape units with its ten, it is unbuffered and,
in general, the additional $1,000 a month for a mc is profitable. This
probably means two will be ordered to provide for a total of 16 tapes,
because very few installations can successfully operate with fewer than
ten or twelve .. With two IRCs on hand, the discussion so far makes it
evident that a double-buffered 1Imicrosecond interrogation" process does
almost double production compared with one IRC. We shall subsequently
consider some of the additional factors (1187) which serve to make the
actual expected increase somewhat less.
1186 .. 4
Double-Buffered 'tRead Into Memory·' Processing.. If all master file tape
records are dumped into the 705 memory, net tape speed is reduced compared
with "microsecond interrogation" and more internal computing time is involved in tape handling. As a compensating factor, much of the actual
productive processing may now overlap tape time, rather than being outside ito The maximum transfer rate is achieved with constant 1,022-character records, which will be used for evaluation purposes because it
gives the most effective 705 performance"
Time for reading one input and writing one output grouped
85 .. 7 ms
record (1,022 characters each) "
006
Non-overlapping between-record computation (minimum) .. .. .. .. •
TRC-to-memory transfer (non-overlapping)
~
95 .. 5 ms
Total time per tape grouped record .. "
Handling another set of tapes on a second mc can be accomplished in the
same time; this is at a rate of 2,044 characters ever 9565 ms, or 1,000
characters in 4608 ms.. Of the 9505 ms~ the following are the minimum required for essential operations not related to effective computation:
0
"
"
0
0
0
0
0
0
0
'"
..
..
0
..
..
..
..
0
mC-to-memory transfers and back (two mcs at 1804 ms each) ... 36.8 ms
Non-overlapping between-records computation (minimum)
104
Data transfers to reassemble inputs into new outputs, not
including loop housekeeping (minimum)
0
705
Total time per pair of tapes
co
..
..
4507 ms
0
0
0
..
"
0
Original 1/31/57
1186.4
(Continued)
Thus it appears that about 50 ms are available for testing records to
determine if they are required for processing and for actual computations.
This is usually sufficient to permit unchecked computations~ in normal
types of processing, to overlap completely the input-output tape operations. Let us now examine the validity of this apparently available 50 ms.
1186.41
Implications of Block Reassembly. The times cited assume blocks of a
constant 1,022 characters and will not be significantly less if the actual
blocks are close to 1,000. To account for additions and deletions, this
means reassembly of an input block into two output blocks as a normal occurrence. Normally, then, an output block is filled with some of the input, and the remainder of the input is carried over to the next following
output.. Because there is only one dual-purpose buffer in the IRC, the
next input block must be unloaded into memory before the output can be
writteno And, as noted in 1184.11, this means either two input blocks
and one output, or two outputs and one input, must be set up. The only
alternative is using one IRC only for writing~ or only for reading, at
a great increase in tape time. The handling of two inputs (or outputs)
means two areas in memory, and this requires repetitive address modification shifting back and forth betwee~ the two areaso (The solution of
transferring inputs to a common work area is possible, but may require
excessive memory transfer time in the 705). The computing time necessary
to do this address modification may vary considerably.. It may be almost
none if separate instruction sequences can be used for each area, but
this is achieved only at the cost of memory space for non-productive instructionso If the memory space is to be freed for useful instructions,
then address modification time is sacrificedo The latter choice, which
ove,rall is probably preferable, may add anything from one ms up to several,
depending upon other parts of the program which may be affected--as an
example, exit points in processing routines may require changing. Something like 2.0 ms, which allows a maximum of about seven changes, is not
an extreme expectacyo With a total of two IRCs affected, this adds up
to about 400 ms as a practical estimateo
There still remains the housekeeping in the individual reassembly program
sequences. As noted in 1184.11, this also can be quite variable, but assuming it will take as much time as the actual memory transfers is, if
anything, somewhat conservative. This, then, is another 3,,75 ms for each
IRC or 7,,5 ms total ..
Thus we have arrived at about 11,,5 ms
time, reducing the time available for
alternative to doing block reassembly
interrogation;·' this immediately adds
length tape records o
1186.42
to be added to the block reassembly
computing from 49.8 to 3803 ms. The
is the approach of "microsecond
to tape time because of variable-
Testing of Master File Items. Unless the high-order index number in each
tape block is in a known 10cation 9 or is repeated at the front of the
block, each item must be examined to determine if a transaction applies
to ito Such a test requires, at a minimum, two instructions per master
item; the time required depends upon the number of characters, but a
representative figure is 00221 ms for an 8-digit index number. If the
Original 1/31/57
1186.42
1186.42
(Continued)
file averages five records per tape block, another 1.1 ms is consumed in
direct comparisons and at least an equal amount in address modification
(part of the loop testing can be combined with block reassembly). This
totals 4.4 ms for two IRes, with a probable range of around 3.0 ms as a
minimum up to more than 10.0 ms for a longer index number and more items
per block. And if the user desires an accurate comparison--which may be
made mandatory by an outside auditor ora governmental regulatory body-multiply these t~es by anything from a minimum of two to more than five.
Any significant reduction in this time presupposes that the high order
index number in each tape record is in .a known location--which means all
items are of constant length--or is repeated at the beginning of the
block, as in t·microsecond interrogation. tt There is nothing to gain in
this latter method in the "reading into memorytt mode, because block reassembly would require its recalculation on each processing cycle.
Using an average of 5.0 ms for unchecked master item testing against a
transaction input, the time available for computing is now reduced to
33 3 ms. If the customer is old-fashioned enough to want accuracy in his
processing, take off another 'ten to thirty ms.
0
1186.43
Control Sequence. The control sequence and tape handling instructions
suggested by IBM are quite short, quite fast, and quite full of holes.
For one thing, they are concerned only with reading and writing the master
file (so far, this analysis has been limited to that phase). By the time
necessary provisions have been made for reading and writing other inputs
and outputs, the control sequence may be a good deal more complicated.
Likewise, some customers are going to want--or the auditor is going to
insist on--safeguards against completely losing a master file block--and
using IBM suggested tape-handling sequences makes this possibility a
certainty. Instead of using 0 .. 6 ms of non-overlapping computation, it
may be anything from 1.0 ms up. A reasonable minimum is on the order of
1.5 ms. This increases the total tape time by 0.9 ms (the difference between the 0.6 ms and the 1.5 ms now used) to 96.4 ms, and reduces the time
available to 32.4 ms.
1186.44
Productive Processing. It would appear that this 32 4 ms is enough for
most normal processing without throwing the 705 into a computer-limited
status.. This figure is, of course, only a rough average; the amount of
time may range from nothing up to possibly 45 ms with 1,022-character tape
records, enough for around 225-450 instructions. This seems ample.
0
Now let us consider the two basic ways of doing necessary actions on a
master file being processed in two parallel sections. Each has itsown
input areas, and, similarly, each has its own input area to hold the next
detail or transaction item awaiting processing when the matching master
shows up.
The first method is to set up individual programs for each 'IRC, identical
insofar as what is done is concerned, but having addresses tailored to
each of the two sets of data storage areas.. The functions depicted in
1186021 are then connected in this manner:
Original 1/31/57
1186 44
0
( 1)
1186.44
(Continued)
Processing
Routine A
Control
Sequence
Processing
Routine B
This is quite straightforward; as soon as the control sequence finds one
IRC ready with its next record, it transfers to the input-output sequence
for that IRC. This sequence handles the tape reading and writing, examines the new input to see if any item requires processing, does whatever
is required when a "matchtt occurs, reassembles the input into a new output and then--when this complete cycle is finished--transfers back to the
control sequence. Therevs only one thing wrong with it: Except for the
control sequence and program constants, everything in memory is duplicated.
Compare with a single.-buffer type operation,. this approach can accommodate well less than half as many productive instructions. It's well
less than half, rather than about half, because both instruction space
and input-output space for the second IRC would be available for more useful instructions if only a single set of input-output tapes were being
handled ..
So, although straightforward, this approach often must be discarded because
it seriously restricts the amount of work which can be done in a single
pass of the data through the 705. However, it should be noted that this
is the precise method followed if two different jobs are being multiplexed
on the 705.
The second method is to use a common processing routine for handling items
from either IRC; there are still separate input-output sequences for
each IRC (probably including the master item tests), but both of these
transfers to a common program for any productive computation required. The'
graphic relationship is this:
Control
Sequence
Common
Processing
Routine
.--
Original 1/31/51
1186.44
(2)
1186.44
(Continued)
It appears that this uses memory more effectively, because now a program
twice as long as the first method can be contained. Actually, it is not
all gain, because more housekeeping is required. Obvously, the exit from
the common routine must be set to the proper input-output sequence before
it is entered. This is not enough. Because the inputs can be in two different areas, it is necessary either to move them to a common working
storage area or to do considerable address modification within the common
routine to adapt it to the particular IRC involved. In most practical
cases, the former choice is probably preferable, but this may not always
be true. It is, for example, the conventional method used in Univac processing.
In the double-IRC case, it is necessary to move both the master file and
transaction input items to the common working area. Although the so-called
high-speed memory transfer most likely can be used, it still takes time.
Moving a 200-character master and a 40-character transaction item, plus
setting the exit from the common routine, take a minimum of 1.15 ms after
the transfer addresses have been set. Likewise, the time required to restore the updated master item into-its proper place in the new output always is greater than a straight input-to-output transfer.
Thus it is seen that the normal method of using a common processing routine
takes some time in housekeeping, but not a damaging percentage of available computing time. Apparently, most processing can still be done within
tape time and a double-buffered 705 can process master files at a rate df
1,000 characters about every 47ms.
A timing chart becomes useful for further analysis.
picture:
Tnis is the basic
IRC A tape handling
and testing sequence
)
(
3
Write and Read with IRC A
~
~
)
Control Sequence &
Available for Comp.
~
~
Write & Read with IRC B
CB
Assume this is a base condition, with IRC B leading IRC A by a time just
sufficient to permit completion of all IRC A tape handling, master file
testing and block reassembly by the time IRC B is ready with its next
tape record. Now if IRC A has an item requiring processing, the movement
of its tapes can continue without interruption, but IRC B becomes
t'stalled tt while the computation is being completed.. It then gets out of
"synchronism t • with IRC A and the tendency is to stay that way.. The effect, then, is similar to that encountered in ttmicrosecond interrogation: t '
The two IRCs interefere with each other, and an "interference allowance"
must be added to the tape passing time o
Original 1/31/57
1186.44
(3)
1186.44
(Continued)
The question naturally arises: "Can the sequencing of the programming
be so arranged that this interference is avoided?" The obvious solution
is to do the tape handling in two successive operations, and perform all
other operations after tape instructions are out of the way; the graphic
chart may be considered in this fashion:
Reassembly
and Test
IRC
Processing
IRC
A
B
Processing
IRe
B
Here the ttcontrol sequence tt vanishes in a straight-line operation; IRC A
is always handled before IRC B. Both new input records are moved into
memory, both outputs moved into the IRCs, both tape writing and reading
operations are initiated and then the functions of master item testing,
any processing required, and block reassembly are donem So long as these
latter computations do not extend beyond the period when IRC A is ready
with its next ~ecord, internal processing can apparently overlap completely
the tape movement from both TRCso This is exactly the objective of the
double-buffer process and is precisely what IBM is claiming as possible
with two IRCs. Moreover, it gives an effective tape handling rate of
1,000 characters about every 48 ms, definitely faster than Univacvs 720
every 51 ms.
Well, what--if anything--is wrong with it? It seems to be quite logical
and to cover the normal processing requirementsc But it noneth~less has
not one, but several, loopholes which we will now take a look at.
Only an engineer would think of the first one--but hevd think of it immediately. The example has postulated tape records of fixed lengths--l,022
characters-7as being most favorable to the 705. Two of these are being
read simultaneously and the peak performance noted above assumes that
reading two records of identical length takes an equal amount of time.
Mechanically, this is almost impossible to achieve; it demands absolute
uniformity in the speed of tapes during both the writing and reading operationsc Even with perfect maintenance, this uniform time of reading cannot
be achieved in practical operationo The tape passing time of both IRCs,
under the processing method above, is slowed down to the speed of the
slower input tapeo How much is it? Well, one unit 3% slower than the
other adds more than 1.0 ms per 1,000 characters, and a variation from
rated speed of this amount is probably more common than a lesser one.
Not much, perhaps, but EDPM operations are replete with one millisecond
figures--which have a nasty habit of cropping up several places in a program and suddenly adding minutes to a machine run.
The second weakness in this approach is that it requires constant-length
Original 1/31/57
1186.44
(4)
1186.44
(Continued)
tape records, which with the 705 almost inevitably means a fixed-length
master item. It has already been pointed out (1184.2) that typical business files consist of items inherently variable in length and forCing them
into a fixed-length pattern means one (or both) of two things: (1) ttpadingu with blanks to fill out shorter items to the standard length, or
(2) elimination of the variable elements from','the file and putting them
on a separate tapeo The first increases tape-handling time; the second
either generates another machine run or moves the bothersome variable
elements to another tape unit--where they still must be taken into account and read into the machine in some mannero
Establishing a file in variable-length format immediately rules out the
practicability of the processing in this last manner and makes a tlcontrol
sequence t • mandatory for most efficient operation. And--at the same time
--it introduces the variable record length allowance (as in "microsecond
interrogationtt) and an interference factor between the two IRCs. The
latter factor is of more importance than it was in t'microsecond interrogation" for two reasons: (1) There is even less computing time available for productive processing because more is taken up in housekeeping,
and (2) the percentage of transactions usually is higher, resulting in
more chances for interference. (Even IBM does not recommend ttmicrosecond
interrogation" for high-activity files). The variable record length allowance may be somewhat less than in the other mode, because block reassembly can eliminate the extremely short records. But at the same time,
the internal housekeeping instructions necessary to use the high-speed
memory transfer in block reassembly now take more timeo In general, it
may be considered that the combined effects of increased reading time
because of variable-length tape records and the interference between IRCs
will add a minimum of around 15% to the nominal tape time 6 (using a 500 or 1,()OO line printer) or high-density
printing requirement (~sing.a 774-407 printer in the latter case). The
elements printed go on one tape, in completely edited format if necessary,
and those not printed on the other. The lack of a printer control panel
on the high-speed printers, rather than processing considerations, is the
major reason for this approach.
From the standpoint of file maintenance, splitting a master file into two
(or more) sections introduces several weaknesses in addition to those already discussed under the two basic modes of operationo Because the two
sets of tapes must be processed in complete synchronism, both parts . of
each item must contain an identifying index number. This duplication adds
to the total number of characters in the file and this adds to tape-handling
time; it also requires otherwise unnecessary internal comparisons to
make sure that corresponding records are being processed.
Both sets of tapes now move record by record, rather than independentlyo
This means that both are limited to the speed of the slower (probably the
one with the most information); it is a rare occurrence when both parts of
such a split file are of approximately the same number of characters.
A further restriction on tape speed arises when requirements of a printing tape limit the maximum block to 600 characters; this condition arises
whenever the print line exceeds 98 characters. A tape block of this size
is read about 15% slower than one of around 1,000 characters insofar as
master file time is concerned.
The processing rate, in time per 1,000 characters, evidently is subject
to considerable variation.. If encountered in any sales proposal, .an
approximate time or rate can be determined quite readily by using the
principles and factors discussed in the previous sections. In general,
something on the order of 70-75 ms per 1,000 characters is considered to
be a rather better-than-average expectation.
1187
1187
SYSTEMS FACTORS AFFECTING DOUBLE-BUFFERED PROCESSING
The discussion of previous sections has been limited largely to factors
affecting the processing rate of the master file. Their basic objective
has been to explore program and operating facts which derive from the
705's particular logic and to point out practical considerations which
must be taken into account in evaluating double-buffered file processing
under either major mode of IRC operation. Some mention, but only cursory,
has been made of other factors which affect the time and cost of this
type processing. These are examined more fully in this section.
1187.1 Tape Units and Number of Input-Output Files. The typical EDPM run involves more than merely feeding a master file and a transaction file
through the machine to arrive at an updated master. There may be other
inputs and practically always other outputs exist. Any proposed EDPM
processing scheme must provide for handling all necessary inputs and generating all required outputs.
In a sequential magnetic tape processor, all inputs must be in the sequence of the master file. (The fact that the 705 may have a magnetic
drum, which might permit handling small volumes of data not in master file
sequences, is not considered in this section. Refer to section 2520 for a
discussion of its use). Outputs likewise are generated in this sequence,
although sorting or other types of additional work may be necessary to
put them into the proper order for a final report or auxiliary tape file
record. In general, some outputs are to be used in the master file sequence, while others may require sorting into a new sequence.
1187.11 Necessary Duplication of Tape Units. In double-buffered processing of a
master file, duplication of tape unit assignment is inevitable. In the
typical single-buffer EDPM run--both Univac and 705--four tape units are
assigned for handling the master file, two input and two output. This
method permits continued reading or writing with one unit while the other
is rewinding and being changed. In addition, either one or two tape units
are assigned to the transaction file; for simplicity, assume it is short
compared with the master file and that one is enough. Five tape units,
then, are. required for these two files.
In a double-buffered process, five are needed for each IRC, or a total
of ten. The master file uses four of them, and a separate transaction
tape is required for each half of the master file. (It is easy to show
that a single transaction tape would not work)e
This is the minimum number which can be assigned. If a proposal involves
less, it means that processing from one IRC will be suspended during rewind and tape changing, on both input and output tapes, which usually
will not terminate simultaneously (although, in some instances, they can
be so set up). Processing of the tape with the other IRC could continue
during this period, but would require additional programming in the 705.
If either the input or output is limited to one tape unit instead of two,
master file handling time is increased a minimum of about 500.£--about three
minutes rewind and tape changing time are added to the 6.4 minutes required
to read an average full reel of tape in a tape-limited operation. If both
are limited to one tape unit, master file handling time is about doubled.
Original 1/31/57
1187.12
1187.12 Other Inputs and Outputs. Tape units are also needed for other inputs and
outputs. Inputs must always be assigned in pairs, one for each IRe. Outputs need not be, because a subsequent machine run can separate or sort
them as required. Let us consider" the two basic types of output and their
effect on a double-buffer operation.
1187.121 Outputs in Sequence of Master File. Some outputs should be in the same
sequence as the master file; examples are transaction register tapes for
printing, exception on error item tapes, and transaction history tapes.
In a single-IRe 705 or a Univac II run, such outputs are obtained automatically sequenced during the normal processing. In a double-IRe process,
they are available in correct sequence only if two tape units are assigned.
This, of course, costs money--and uses up tape units pretty fast.
If a single tape unit is assigned for such outputs, there are two adverse
effects. The immediately obvious one is that a separate machine run is
required to separate the interspersed items onto two tapes; this of
course reduces the ostensible time savings of a double-buffered process.
Second, the fact that the entire output is handled through one IRe tends
to increase the number of characters it must transfer as compared with the
other one, and this again increases the time for master file processing.
This can be offset, at least partially, by using each IRe for one such output, but seldom will they be of a nature or volume requiring the same time
to handle.
One other facet of the me method of operation also becomes evident at
this ;~point. Univac has a separate input and output buffer which never
interfere with each other. The Type 777 uses one buffer for both purposes; this means that every output instruction must recognize the possibility of an input block in the buffer and provide for dumping it into
memory--in its proper place--before the output block is moved. Often this
can be highly inconvenient; it is just possible that there is no convenient
place available for it. Similarly, master file tape handling orders must
recognize the possibility of there being no input block in the 'IRe at the
time a read-write order normally would be executed. The net effect is
that tape control program sequences using either one or two 'IRes are more
complicated than those of Univac and increase tape handling instruction
time above the figures arrived at by using IEMus oversimplified sample
routines.
The dual use of the IRe buffer is responsible for one other big difference
between 705 and Univac II programs. In Univac, input and output instructions are completely independent of each other; after a subsidiary output instruction is underway, the program can shop around at leisure to see
if another input can be used, or can read in another master file block.
With the IRe, this determination must be made before the output is disposed
of. Once it becomes engaged in either a single read or a single write,
it is completely tied up ....1 the tape-buffer transfer is finished. There
is with the 705 a much greater probability of single-tape operation than
with Univac, which can use either buffer without regard to the status of
the other.
Original 1/31/57
1187.122
1187.122 Outputs Not in Sequence of Master File. Some outputs, which are of
course generated sequentially in the file processing, must be sorted into
a different sequence for other purposes. For such outputs, there is no
basic loss in placing items from both IRes onto one tape. As in the
previous case, some loss of time most likely will occur because one IRe
must handle more characters than the others, and the comments of the last
paragraph are applicable here too.
1187.2
Duplication in Memory. It has already been noted that the double-buffer
approach requires duplication of memory areas for both data and for instructions. This may be relatively small; it could be only a few hundred
extra character positions in "microsecond interrogation." In the "reading into memory" mode it involves several thousand, or a significant percentage of the space available. If a master file is set up in tape,records
of 1,000 characters or so, each IRC requires an input-output area of about
3,000 character positions, or 3~ of the total memory--15%" for each IRC.
More duplication occurs in the transaction input area, although this could
be held to single item size at some sacrifice in input tape speed (and
probably a considerable sacrifice in the final merge routine that sorted
the input items). More duplication occurs in the two sets of tape handling instructions and block reassembly sequences, and still additional
requirements are imposed by the control sequence and special instructions
necessary to set program "switches" and otherwise provide for conditions
not met with in single-buffer processing. At a minimum, a double-buffer
705 operation will cost at least 4,000 character positions in duplicated
functions--2~fo of memory.
From 5-7,000 is a more normal expectancy. Obviously, this means that less than usual can be done in one machine run;
there simply is no way around the fact that a given area in memory can be
used for only one thing at a time--and 5,000 characters of duplicated
memory is equivalent to 1,000 instructions. Even if 705 memory is expanded to 40,000 characters (at a monthly cost of $2,500 more), a doubleIRC run means something between 800-1200 fewer productive instructions.
(Refer to 2520 for comments on use of a drum for storing instructions
which exceed memory capacity).
1187.3
Non-Overlapping Operations. A surprlSlng number of mundane operating
requirements can stop an EDPM process completely. We have already mentioned
some practical factors which will stallone IRe and thus prevent full movement of both sets of tapes. These basically are stoppages which arise because of the logically necessary implications of a double-buffer operation and some of the characteristics of the 705/777. This section points
out still more factors which enter into the picture.
1187.31
Machine and Data Transfer Errorso Every m~chine error or tape readwrite error causes a temporary machine stoppage or rerun which affects
both TRes. Even failure to read a tape block on the firsttry--a common
occurrence on IBNP s Type 727 Tape Uni ts--causes a quarter-second delay
with average size blocks--and it would be a beaut of a tape re-read subroutine which would keep the other IRe moving during the correction.
Similarly, any restart procedure .initiated by a machine error, of any
type, consumes considerable time in ll'epositioning the tapes to the proper
pointe Because IBM tapes cannot be read backwards, they are conventionally rewound and then individually ,advanced to the proper position. This
may take ten or fifteen minutese Of course, such a point may not be '
reached for qUite a few hours, but also may occur two or three times
within a single hour. Restarts may average once every two or three hours.
Original 1/31/57
1187 .. 31
1187.31
(Continued)
to once in ten or twelve, depending upon the specific 705; for some reason,
EDPMs seem to have their own personalities and two successive units off
the production line can (and usually do) have widely different ttquirks."
The slowdown effect of machine errors is extremely difficult to predict,
but is definitely significant. For example, ten minutes delay in a tenhour machine run sounds nominal, but adds almost one millisecond to the
time necessary to handle 1,000 characterso And ten minutes is hardly
enough to correct one machine error with a programmed restart procedure.
And one machine error in ten hours is a rate even less thanIBM admits
exists; they acknowledge that one completely unreadable tape error occurs
on the average every eight hours. The actual incidence of this specific
type error is known to be considerably higher in several installations.
1187.32 Check Points. Standard EDPM operating procedure calls for periodic writeouts of memory on a "scratch tape tt as a part of program-controlled restart
procedures. Although the frequency varies with different installations,
two minutes is a recommended interval (oftener than with Univac, by way
of note) and one adopted by some 702/705 installations for which the period is known. LetVs forget the fact that programming and computing time,
usually is some form of block count, are necessary to measure the interval
between successive check points.. Just count the time required to write
out 20,000 characters, backspace the tape and read the write-out to verify
its accuracy. About 1.4 seconds for each phase, or 402 secondso Doing
this every two minutes adds about 1.7 ms to the per-l,OOO character reading rate. Remember--nothing else occurs during this period.
1187.33
Tvpewriter Printouts. The typical EDPM program provides for considerable
chit-chat between the machine and the operator; instructions to change
tape reels, announcements of machine or tape error subroutines, etc .. , keep
the operator from getting lonesome.. On Univac, most of these are one-word
typeouts which don't slow down processing; a one-word buffer permits
Univac to overlap computation and typing. The 705 comes to a halt during
such printouts, because its typewriter is unbuffered ..
This apparently minor difference can have an appreciable effect on tape
handling times.. For example, in a double-lRC 705 operation, the typeouts
to change master file tape reels occur at a nominal rate of four every
six to seven minutes. If they took only one second each--shorter than the
conventional reel""changing instruction--the four require 1% of total reel
time, adding 0.5 ms to the per 1,000 character rate. And this is only
one of several types of printouts which may occur on a routine or special
basis. Just to add another example, consider the typeout for tape labels
of new output reels--at four or five seconds each.
1187.34 Machine Failure.. Although we should never mention to the customer the
fact that the machine fails, the mere presence of maintenance engineers
is a sufficient remindere It goes without amplification that any maintenance required on a double-buffered operation stops two sets of tape
units rather than just one, and has a considerably greater effect on the
amount of production per dollar ..
Original 1/31/57
1187.4
1187.4
Summary of Double-Buffer Processing Speedup. The detailed analysis which
has been developed makes it readily apparent that a double-IRe operation
is never going to result in doubling the processing rate of a file, regardless of how much IBM preaches this magnitude of gain. Of a given
machine run, part of the total time represents work which can be split
between two mes at a definite time improvement, but nothing like a 5CJ',£
reduction; the remainder is time that is not eliminated or reduced. For
example, a 10-hour machine run might involve 7 hours of master file work
plus 3 hours of handling subsidiary files, error correction routines and
printouts. To process this file on two IRes might reduce the 7 hours to
something like 4 or slightly more; the 3 would remain essentially intact.
The net reduction, then, is from ten to seven hours, or 3CJ',£,not 5CJ',£. In
general, a double-IRe operation may be considered as capable of saving from
about 25% to 4CJ',£ of processing time, the actual saving depending upon the
cumUlative effects of the numerous factors which have been discussed in
this and previous sections. This is equivalent to increasing the amount
of productive work per hour by from 3~,£ to about 7(1',£, but nothing close
to doubling it. When all factors are taken into account, it does not appear that an increase of more than this percentage (70%) can be realized
in practice.
1187.5 Multiplexing of Two Different Jobs. Previous sections have paid a good
deal of attention to factors affecting the processing of a large master
file by multiplexing it through two TRes. It is also possible to handle
two completely different jobs, one on each me, and IBM has advocated this
type of approach alsos (For a rather obvious reason--many installations
spend only a small part of their time on files big enough to justify
spIrt-me processing)e Little has been said about the requirements which
must be met before this multiplexed processing is practicable. The more
important are listed here (not necessarily in order of importance).
1.
Both jobs must take about the same machine time when performed singly.
This is obvious.
2.
Both must be definitely tape-limited when performed singly. If
either is computer-limited, or close to it, there is no gain in multiplexing them ..
3.
The combined number of tape units must not exceed the number in the
installation~
More specifically, each should be self-contained within. the maximum of eight which one IRe can control, although this is
not mandatoryo
4.
The combined program and data storage areas of the two jobs must not
exceed the capacity
memory. This is either 20,000 or 40,000 characters.
5.
Both jobs should be handled on the same processing cycle. Actually,
one could be done singly on a daily cycle and the other multiplexed
with it only one day a week; evidently, any savings accrue only when
multiplexing is donee
of
All comments previously made apply in this case. In general, it appears
that interference between IRes will be of more significance, although this
may not always be true.
Original 1/31/57
1188
1188
COMPARISON OF UNIVAC II WITH DOUBLE-IRC TYPE 705
Except in cases where IBM proposes a minimum 705 installation to try to
beat its cost down to that of a Univac I, it can be expected that most
proposals in competition with Univac II will involve two, or even three,
IRCs. This is inevitable, because two are necessary to provide buffered
control facilities for more than eight tape units and, in addition, a 705
with only one IRC inevitably loses out to Univac II simply on ability to
transfer information to and from tapes. The real reason competition will
advocate double-buffered operation is simply that it is possible anyway
if a workable number of tapes is included in an installation. It is thus
pointless to try to convince a prospect that a double-IRC operation is not
profitable; if he is going to get a 705, he will definitely profit by
multiplexing all jobs he possibly can. Our approach can be much more
direct: In jobs not multiplexed, Univac II will out-perform a 705 by a
minimum of 35% in speed and 3g( in cost-of-performance. If multiplexing
is involved, Univac II may be somewhat slower, but will match the 705 in
cost-of-performance under optimum conditions for the latter. Because optimum conditions cannot always exist, Univac II will always give acrossthe-board superior and more economical performance than the 705.
1188.1
Costs of Typical EDFM Installations. Because the cost c;omparisons to be
developed involve only the main frame and tape units, nothing else need
be taken into accounto Monthly rental figures are shown.
Number of
Tape Units
6
7
8
9
10
11
12
13
14
15
16
Univac II
IBM Type 705 (20,000 Cha:tacters)
One IRC
$ 21,240
21,690
22,140
22,590
23,040
23,490
23,940
24,390
24,840
25,290
25,740
$20,300
20,850
21,400
Three IRCs
Two IRCs
$ 23,300
23,850
24,400
24,950
25,500
26,050
26,600
27,150
27,700
28,250
28,800
$
28,500
29,050
29,600
30,150
30,700
31,250
31,800
The cost comparisons which are developed assume that both Univac II and
IBM 705 will use instructions tape, rather than program cards for the
latter equipment. This is the cheaper way for the 705 to operate, because a card reader costs several times as much as a tape unit.
1188.2 Comparable Installations, Univac II and IBM 705. Equipment configurations
providing essentially the same processing capabilities depend upon how
the IRCs axe used.
1188.21 Two IRCs With One Input and One Output. If one IRC is used only for in-~
put and the other only for output (i.e., no multiplexing), the 705/777
requirements for tape units are essentially the same as for Univac. The
basic difference is that each IRC can control only eight tape units, which
might make difficult holding to using each for only a single purpose.
Oxiginal 1/31/57
1188.21
1188.21
(Continued)
In this type of operation, comparative equipment costs can be obtained
from the Univac II and second 705 columns in 1188.1. Note that Univac II
always costs less than a comparable 705 installation.
1188.22
Two IRCs With Multiplexing. When complete multiplexing is involved, a
705 installation will always have more tape units than a comparable Univac II. This is true if the major use of multiplexing is processing a
large master file in split sections, or is parallel handling of two different jobs in one machine run. The former is much the more common case,
and requires a minimum of five tape units more than a Univac II doing the
same job. This is based upon four for the master file and one for the
transaction input. Costs of comparable installations then are:
Univac II
Uniservos
Monthl::! Rental
6
$ 21,240
7
21,690
8
22,140
9
22,590
10
23,040
11
23,490
An IBM lype 705 installation
Univac II.
1188.3
IBM 705 Wi th Two IRCs
Tape Units
Monthl::! Rental
$ 26,050
11
12
26,600
13
27,150
27,700
14
28,250
15
28,800
16
costs roughl::! 25% more than a comparable
Cost of Performance, Single Buffer Tvpe Operation. If the 705 does not
use each IRC for both reading and writing, the relative processing rates
of the two equipments are directly comparable. In a tape-limited operation, Univac II will pass a master file at a rate of 720 characters every
51 ms, equivalent to 70.8 ms per 1,000 characters. The 705 will pass a
1,000-character record about every 82.2 ms, reading in mode B2. The relative cost of performance for varying numbers of tape units may then be
calculated, using the Univac II cost in each case as 100:
Number of
Tape Units
8
9
10
11
12
13
14
15
16
Monthl::! Rentals
IBM 705
Univac II
$ 22,140
22,590
23,040
23,490
23,940
24,390
24,840
25,290
25,740
$24,400
24,950
25,500
26,050
26,600
27,150
27,700
28,250
28,800
705 Cost of Performance
127.9
128.2
128.4
128.6
128.9
129.1
129.3
129.6
129.8
The IBM 705, then, costs 28-30% more than Univac II per unit of production. Further, this is the minimum differential favoring Univac; it applies when9the 705 is processing master files records of constant length,
thus permitting complete parallel operation of the tapes. In the more
typical case of variable length items--and normally variable length tape
records--the 705 time per 1,000 characters goes up, resulting in a still
higher cost-of-performance ratio.
Original 1/31/57
1188.3
1188.3
(Continued)
These figures simply establish factually what is obvious from Univac's
faster input-output rate: The IBM Type 705 is never competitive when
reading and writing single tapes. The sole possible exception is a
highly computer-limited operation, handling factors of very short lengths,
and in which accuracy is either not important or can be verified in other
ways; in this relatively uncommon case (at least in business data processing), the ability of the 705 to operate on short words faster than
Univac II may give it a better cost-of-performance ratio.
1188.4
Cost of Performance, Double-Buffer Type Operations. In a double-IRC
process using each for both reading and writing, the comparative costof-performance depends upon the actual speed of operation attained. Previous sections have discussed in considerable detail the numerous aspects
to be taken into account. Cost comparisons can be summarized by postulating 8 Uniservos with a Univac II and 13 tape units with a 705; this is
not to be construed as meaning that an 8-Uniservo installation is a good
Univac II configuration, but is used simply because one IRC controls that
number. Any upward change to a maximum of 11 Uniservos or 16 Type 727
tape units will cause only a small change in the cost indices developed
below.
1188.41 ttMicrosecond Interrogation. tt The theoretically-maximum transfer rate is
1,000 characters every 42.5 ms; this of course allows no time for processing and assumes that all tape records remain a constant 1,000-plus characters. The more typical maximum expectancy, as discussed in 1186.2, is
at a rate of about 1,000 characters every 55-56 ms. When some of the
non-overlapping operations discussed in 1187 are considered, a rate of
1,000 characters every 60 ms appears to be an optimistic production hope.
Univac II does not operate in this mode, but handles characters at the
rate of 1,000 every 70.8 ms. The relative costs of performance may be
calculated for several time rates of the 705, again using Univac II as
100.
705 Time per
1,000 Characters
Relative Cost
of Performance
42.5 ms
50.0
55.0
56.0
60.0
65.0
7000
73,,6
86.6
95.3
97 .. 0
104.0
112.7
121.2
Thus it will be observed that under the optimum normal conditions, the
705 using two IRCs in itmicrosecond interrogation" is barely better than
Univac II in cost of performance. In the more normal condition, with
60-65 ms rates per thousand characters, it is somewhat more, the differential becoming appreciable for a 65 ms rate. In practice, the difference
is even more marked because of the introduction of other factors--such as
"paddingft of tape records--which virtually negate the applicability of
this mode of operation.
Original 1/31/57
1188.42
1188.42
"Reading Into Memory." The maximum 705 character transfer rate with two
TRCs reading into memory is 1,000 every 48 ms, with a normal maximum of
1,000 every 55-56 mSa Like "microsecond interrogation," this is probably
a better rate than can be achieved in practice, and something between
60-65 ms is considered the optimum working rate which can be expected.
Because the tape unit complement is identical with the case previous, the
relative cost-of-performance figures are also the same. Thus, under
the best possible conditions, the 705 will have a slightly better cost-ofperformance ratio than Univac II; under the typical working conditions,
it will range from slightly to definitely more expensive~
1188.5
Summary of Comparative Costs~ The determination of Univac II time for a
given operation is much simpler than a double-buffered 705; in our case,
it is not necessary to take into account such definite--but indeterminable-factors as possible interference between the two sets of tapes and variations in tape record lengthso IBM can be expected to propose methods most
favorable to double-buffered processing; with the basic logic of the 705,
this .almost inevitably means an emphasis on fixed-length master items (a
punch-card approach that has seldom been pronounced in Univac applications), and this, in turn, entails at least some padding of records. The
discussion of previous sections has gone into considerable detail on the
ramifications in the two basic modes of operations of a two-IRe 705, with
the significant points in both file format, tape handling and normal
operation developed. The net conclusion is that under favorable circumstances, the two IRe 705 is approximately cost-competitive with Univac II
and in more normalci~cumstances is definitely more expensive.
It must be remembered that not all operations are susceptible to this type
of treatment. IBM itself does not make this claim. And when the two
IRes cannot be used for reading and writing in each of them, Univac is
always faster and cheaper. And accuratee It takes very little time per
8-hour shift before any savings made on a two-IRe operation vanish. For
example, if the 705 can do double-buffered work at a rate of 1,000 characters every 50 ms--an exceptional rate--on some files and only singlebuffer type operations the rest of the time, it would require 5t hours
per 8-hour shift of double-buffered operation before the 705 became,
overall, competitive with Univac 110 In the normal optimum expectancy,
with a cost-of-performance margin of 3-5% over Univac II as maximum, a
half-hour a day of other work would more than offset any cost savings
which might accrue from the double-buffered period. And in almost any
imaginable application? more than that time would be spent in single-type
work occurring during a double-buffered machine rune
The net conclusion is that a double-buffered 705 is rarely cost-competitive with Univac II; the margin will range from a minimum of l~fo up to
more than 25% for the Univaco And Univacvs accuracy is a built-in and
free bonus to this better cost-of-performance ratioe
Original 1/31/57
1189
1189
SALES POINTS ON TYPE 777 TAPE RECORD COORDINATOR
Many of the points included in this section refer to the detailed discussions of 1180-1188. A complete evaluation of the Type 777 IRe can become extremely complex; however, all of the essential factors have been
explained and analyzed in detail and should be referred to as necessary
in combatting IBM claims for the performance of this unit.
1189.01 t'The 777 Doubles the Speed of the Computer. i' IBJIA has been advertising
the IRe as doubling the speed of an EDPM by permitting simultaneous readwrite and also overlapping computation ..
For the 777: True, with some restrictions.
Against the 777: Well, it depends upon the reference point. Insofar
as IBM is concerned, the statement is true. Of course, Univac I had
the facility in 1951, still has it, and probably always will. Simultaneous read-write-compute is not new; it was in the first EDPM eyer
built for business data processing--Univac I. Well, it took IBJIA only
years to almost catch up.
5i
1189.02 Speed-up Possible With Double-IRC Processing. IBM states, in the introductory paragraphs to the Type 777: "If the file is split in half, the
over-all speed of processing should be approximately double. tt
Against the 777:
Never true.
Consider these factors, among others:
(1) Effect of variable-length tape records (1183.2).
(2) Padding records to constant length, in either "microsecond interrogation" or reading into memory .. (1184.1, 1185.4).
(3) IRC interference and non-overlapping computation (1186.2, 1186.3)~
(4) All factors in 1187.
1189.03 Use of "Microsecond Interrogationo 90
in this mode as a time-saver~
IBM may propose maintenance of files
For the 777: The mode is possible ..
Against the 777: These factors work against this mode of operation:
(1) With only a single IRe, use of the unbuffered Type 754 Tape Control Unit is always faster-and cheaper.
(2) With two IRCs, and for basic implications of this method of processing, refer to 1185, 1186.3, 1187.
1189.04 eost Comparisons. See 1188. Note that a single-IRe 705 is never competitive with Univac II and a double-IRC installation always costs more.
1189.05 Detailed Discussions of IRC Operatinq Implications. Refer to the index
preceding Section 1180 for the sections discussing the various types of
operations and their processing implications.
Original 1/31/57
1400
1400
DATAMATIC MODEL 1100 MAGNETIC FILE (TAPE) UNIT
The Model 1100 is the standard tape unit for use with the DATAmatic 1000
data processing system, both for on-line and off-line operations.
1400.1
~.~
Purchase:
Rental:
$60,000
1,350 monthly.
1401
TAPE UNIT AND MAGNETIC TAPE CHARACTERISfrCS
Magnetic Tape and Reel:
Tape base:
Tape
Tape
Tape
Tape
width:
length:
reel diameter:
cost;
Model 1100 Magnetic File Unit:
Number of recording channels:
Mylar; ferric oxide magnetic surface is
laminated between two sheets of mylar
3ft
675, 1350 or 2700 feet
14ft
675 feet
$230, including two reels
ft
ft
tt
1350 feet
400,
it
ft
tt
2700 feet
775
36:
5 control channels
31 information channels
Character recording density:
Approximately 170-175 bitS/inch. (1200
numeric or 800 alphanumeric characters
per linear inch of tape)
Tape speed (forward & rewind): 100 inches/second
Character Transfer rate:
Ins tantaneous,:
120,000 digits or 80,000 Characters/second
Maximum Effective:
60,000 digits or 40,000 Characters/second
Tape start time:
6 ms
Tape stop time:
4 ms (maximum)
Tape reversal time:
Not known, but not instantaneous
Bad spot detection:
Permanent recording in control channel
Detecting physical ends qf tape: Permanent recording in control channel
Method of Recording
Frequency modulated signals
Mode of recording:
31 channels in parallel
Length of tape record (block): Fixed; 62 words, 52 bits eacho Each of
31 channels contains two words
Check bits:
4 bits of each word used for checking
Length of inter-block gap:
Approximately .625 inches
Yes; gaps are filled by recording in
Recording in gap:
backward movement of tape
Tape capacity (2700-foot)
37,200,000 decimal digits
24,800,000 alphanumeric charqcters
3,100,000 DATAroatic words
50,000 DATAmatic blocks
Original 6/1/57
1402
1402
CONTROL UNITS AND USES OF THE MODEL 1100
The va~ious configurations of control units with which the Model 1100 may
be used are listed below, for both on-line and off-line operations.
Refer to write~ups on the various control units (converte~s) for more
detailed information.
Central Processor Operations:
Model 1000 Central Processor
Maximum of 100 tape units
8 tape units
Model 1000 basic unit has controls for
Model 1052 Auxiliary Central Processor has
2 tape units
controls for (additional to basic)
4 tape units
Model 1054, as above, for (additional to basic)
6 tape units
Model 1056, as above, for (additional to basic)
8 tape units
Model 1058, as above, for (additional to basic)
(Models 1052, 1054, 1056 and 1058 apparently may
be combined to provide the control unit$ needed
for the maximum installation of 100 tape units)~
Periphery Operations:
Card-to-Tape Conversion:
Tape-to-Card Conversion:
Tape-to-Printer:
Tape-to-Printer (high-speed):
Card
1100
1100
1100
Reader
- 1300
- 1300
- 1400
-
1200 - 1100
Card Punch
Printer
High-Speed Printer
Original 6/1/57
1403
1403
MAGNETIC TAPE FORMAT WITH THE MODEL 1100
The magnetic tape used with the DATAmatic system is always written and
read in 36 channels of recording, of which 5 are control and 31 are information channelso The tape format is identical, for both on-line and
off-line (periphery) operations.
1403.1
Hses of Control Channelso The specific purpose of each of the five control channels is not certain, but these functions are accomplished:
1.
20
3.
«Block u marks, dividing the tape into usable sections of recording
surface. Bad spots, or areas of the tape with flaws, are rejected
and by-passed by these block marks.
Markers indicating the physical beginning and end of useful recording area on the tape.
Separation or distinguishing of block marks to denote those written
in the forward direction of tape movement from those written in the
backward direction.
A 2700-foot length of tape is divided into 50,000 usable blocks. It is
probable that the five control channels are permanently recorded at the
factory ~nd, in operation, are used only for reading.
1403.2
Diy~n of the Tape Into Blocks.
The division of the tape into p~rma
nently-recorded blocks results in i'gapst. between successive logical blocks
which are the same length as the blocks themselves--about .67 inches, In
the forward direction of tape movement, every other block space is used;
in the reverse direction, the intervening spaces are utilized. Thus a
full reel of DATAmatic tape can be packed solidly with data, except for
the imperfect areas rejected during original inspection.
The interlacing of blocks results in a tape recording depicted in this
illustration (using a 50,000-block tape):
Forward Direction of Tape
Movement:·'----------~)
)
~
28
49971
t
29
49970
(
~(----------
30
49969
(
Ba.ckward Direction of Tape Movement
The control channels on the tape contain recordings which indicate the
physical limits of the tape; these prevent execution of a tape instruction calling for continued movement in the same direction. It is, thus,
not possible to read or write "'off the end of the tape tt
0
Original 6/1/57
1403.3
1403.3
Format of a Block. A block of 62 DATAmatic words, each of 52 bits, is
recorded in 31 information channels on the tape. Each channel contains
two words, recorded serially; a block thus consists of 31 information
channels, each with 104 bits of recording.
Omittiryg the five control channels, the recording of a DATAmatic tape
may be depicted graphically:
Channel 1
Channel 2
Channel 3
Word 1
Word 3
Word 5
Channel 29
Channel 30
Channel 31
Word 57 (52 bits)
Word 59 (52 bits)
Word 61 (52 bits)
(52 bits)
(52 bits)
(52 bits)
Word 2 (52 bits)
Word 4 (52 bits)
Word 6 (52 bits)
Word 58 (52 bits)
Word 60 (52 bits)
Word 62 (52 bits)
The 31 channels are read in parallel; thus, one hit from each of 31 words
is read and assembled in the input buffer. The writing operation is the
converse of thise
1403.4 Method of Handling Tape "Skew .. It Problems engendered because of ttskew" in
the tape--that is, the slight twisting of the tape that keeps all bits
from coming under the heads simultaneously--are discussed in 1109.07.
With a 3-inch wide tape, it would be expected that skew difficulties
would make successful reading almost impossible. In the DATAmatic, this
difficulty is overcome by what is called "self-sprocketing" of each channel.
In other types of tape recording techniques--such as those used in Univac
and the IBM series--there are actually magnetized and unmagnetized "spots"
on tape. Because the gaps between tape blocks are also unmagnetized, it
follows that each character code on tape must have at least one binary ttl,"
and the entire character is read in parallelo Excessive skew rather evidently causes erroneous reading. In the DATAmatic method, there are no
U'nmagnetized spots on the tape. For practical purposes, it can be considered that a binary "Pf is a "big one" and a "O·t a "little one It on the
tape. Every bit position has either a 1tbig" or "little" one, and each
channel has exactly 104 magnetized spots in one block. Thus each channel
can be made t'self-sprocketing, it and the bits assembled into a buffer in'"
dependent of any other channel--it is only necessary to be sure that 104
are read from each of the 31 channels. Essentially, this is the technique
followed in the Model 1100 Tape Unit.
Original 6/1/57
2520
2520
IBM TYPE 734 MAGNETIC DRUM STORAGE UNIT
The Type 734 is a magnetic drum for use in the 702 and 705 EDPMs in an
on-line operation under central processor control.
2520.1
2521
Cost:
---r-
Purchase:
Rental,:
$
2,800 per month
CHARACTERISTICS OF THE TYPE 734
Type of data recording:
Drum capacity:
Division of Information:
Characters per section:
Addressability:
Address range:
Number of drums possible:
Drum size (physical):
Drum rotation speed:
Time per revolution:
Average access (latency) time:
Character transfer rate:
Method of Transfer:
Buffer:
Number of information bands:
Sections per hand:
702/705 character codes
60,000 characters
300 sections
200
Each section addressable
1000-1299 for first drum;
1300-1599 for second drum; etc., to
9700-9999 for 30th drum
Maximum of 30
10.7" diameter by 12.5" long
3750 rpm
16 ms
8 ms
0.04 ms per character (40 microseconds)
Bit-parallel for each character,
serial by character
None
30
10 (interlaced)
Original 5/15/57
2522
2522
METHOD OF OPERATION OF THE TYPE 734 MAGNETIC DRUM
The basic method of operation of the Type 734 is identical in both the
702 and 705 systems, with the minor exception noted in 2522010 The time
required to execute the instruction sequences writing on or reading from
the drum are, of course, different, the 705 being faster.
2522.1
Writing on the Drum. Transfer of data from memory to the drum requires a
two-instruction program sequence:
SELECT
WRITE
xxxx
m
wherexxxx is the drum section address
where m is the memory location of the first
character to be wl'itten.
In executing the write instruction, the EDPM waits until the first character
position of the addre$sed drum section is under the heads; this average
latency time is half a drum revolution, or 8 mso Once this point is
reached, the first (addressed) character in memory is moved to the drum
and thereafter characters are transferred from successively higher memory
locations to the drum at the rate of one every 40 microsecondso The operation is terminated by sensing a group mark (705) or record mark (702) following the last data character in memoryo The group or record mark is not
itself placed on the drum, but is automatically converted to a drum mark
(character code 0 00 0000) which is inserted following the last data character on the drumo
The central processor is t'tied up, fI and can do no other operations, during
the entire time required to complete the transfer from memory to the drum.
This time averages 8
004 N ms, where N is the number of characters (including the group or record mark) to be writteno The time for any given
transfer, of course, is variable, depending upon the relative position of
the addressed section to the heads at the time the t'wri ten instruction is
executed ..
I
2522.2 Reading From the Drym. This is the exact converse of writing, and also
requires a two-instruction sequence:
saECT
READ
xxxx
m
where xxxx is the drum section address
where m is the memory location into which the
first character is to be placed.
Once the addressed section is under the heads, the first character is
transferred into the addressed memory location, and successive characters
from the drum are moved into the next higher memory addresses at the rate
of one every 40 microseconds. The transfer is terminated by sensing the
drum mark, which is automatically converted in a record ma"l'k (702) Or
group mark (705) following the last data character placed into memory.
The average time for executing a read instruction is identical with that
for a write"
2522.3
Number of Characters in a Drum Recordo Although each drum is divided into
300 addressable sections of 200 characters each, records of any length may
be written on Or read from ito The record or group mark in memory terminate.s the writing operation; the drum mark ends the reading" Records of
moxe than 200 charactexs are placed in the drum sections fullowing the one
Original 5/15/57
2522.3
2522~3
(Continued)
addressed, the drum mark following the final data character. Thus, although the drum sections consist of only 200 characters each, any number
of them may be chained together to store records of longer length. The
following comments are pertinent:
1.
The drum section addressed always reads or writes beginning with
its first character position; i.e., it is not possible to begin
reading or writing someplace in the middle of a section.
2.
The drum mark follows the last valid data character, and may appear
in any character position within a section. Thus, a 60-character
record would have a drum mark in the 61st position; a 200-character
record would have a drum mark in the ~ position of the ~
drum section.,
3.
Those character positions within a section following a drum mark
are inaccessible and unusable. Reading or writing always statts
with the first position and always terminates with the drum marko
4.
It is not possible to ttwxi te off the end of the drum." That is, if
the last section of anyone drum is filled with data characters and
more characters (or the record or group mark) still remain to be
read, an error condition exists and an appropriate signal is generatedo This statement is also true·if two or more drums are included
in an installation; it is not possible, for example, to split a
1,000-character record
into the last three sections of one drum and
the first two of the other (this can, of course, be programmed with
two sets of drum write instructions).
The number of characters in any one drum record is completely variable
from a: single character up to one less than drum capacity.
2522.4
Errors in Drum Transfers. Each character transferred to or from the drum
is given a parity bit test; the read-write check indicator is turned on
whenever an invalid bit combination occurs. This indicator can be
set to automatic machine stop or placed under stored-program interrogation and control. An attempt to write off the end of the drum also
turns on this indicator.
Although magnetic drum recording. is the most reliable of any of the magnetic :recording media, it is interesting to note that, in one 702 installation, IBM customer engineers have recommended that programs provide for
immediate re-reading of new records if they are to remain on the drum for
any length of time.
Original 5/15/57
2523
2523
COMPARISON OF UNIVAC II AND IBM 702/705 WITH DRUMS
Because Univac II does not have a drum, a direct comparison between it and
a 702/705 with a drum is not possible. There are unquestionably applications in which the availability of the drum permits a 705 to perform pro ....
.ductive work more cheaply than without one. A single application is not,
of course, a proper criterion. Because anything that can be done with a
drum can also be done with magnetic tapes (in a different manner), the
question reduces to the determination of the system configuration which
provides the most economical overall performance.
At $2,800 a month, a magnetic drum represents an approximate 10-1~
increase in the rental of a typical 702/705 installation (central processor and tape units only); the exact increase varies with the particular
tape and tape control unit setup included. To be economically justified,
a drum should enable the system to turn out least that percentage more
work. The addition of more drums beyond the first requires a proportionately greater increase in the volume of productive output. The compaxison
of the cost-of-performance of a Univac II and a 705 with drums becomes,
then, the determination of the time and cost required to accomplish the
total overall work of a given installation under the two methods. Rather
evidently, this determination cannot be given in specific terms, but is
subject to analysis of each customerVs particular processing requirements;
some general guidelines to assist in making this evaluation can be give.
2531.1
Table Lookup. The term "tablet. conventionally refers to such types as
mathematical values, freight rate classifications, public utility rates,
mileage between cities, loan repayment schedules, etc... As used herein,
it is extended to include also such types as cost distributions and statistical summaries which are prepared for presentation in tabular form.
When related to the sequence of the data generating the lookup requirement, most of these are referred to at random, but some mathematical
tables are referenced sequentially for each iteration of a problem solution.
When stored ona Type 734 drum, a table is normally broken down into
sections each approximating 200 or fewer characters. To be used, these
sections must be "indexed~ in the main memory; that is, it is necessary
to maintain in memory an index showing either the low or high table
entry in each section, together with its drum address. A part of the
memory must be set aside for holding this index, and both memory and computing time are needed for the instructions necessary to search the index
and find the drum location of a particular table value needede> The
searching method most economical of instructions -- sequentially through
the index by a loop -- is unfortunately the most wasteful of computer
time; the method normally most economical of time '-"'binary searching -is probably the most wastef\,.ll of instructions. (Occasionally, the drum
address can be calculated from the entry itself, eliminating: this search).
With·200-character records on the drum, average time to obtain anyone
at random is 16 ms (8 ms average latency plus 8 IDS transfer time for 200
characters), plus the computing time necessary to determine the correct
drum location; this can add anything from another 5 to 20 or more ms.
Original 5/15/57
2523.1
2523.1
(Continued)
The first type of table mentioned above is normally not changed during the
processing, and is subject to reading only. The second type must be continually revised as new data are posted, and require both reading and
writing. With records of 200 characters, this adds another 16 ms, approximately, to the time required to make one reference to the drum.
Without a drum, there are two basic methods of accomplishing on magnetic
tape the equivalent of a table lookup. The first is to store the table
sequentially on tape, and to scan this tape forward or backward, as appropriate, to find the required sectionso The time necessary for this
is basically a function of the size (number of blocks) in the table, but
it always requires considerably more actual time than a drum reference.
Like a drum reference, it requires computation to determine the particular
tape block needed for any specific table value.
The second method is to sort the input data into sequence on the table
index n'umbers and process these sequentially against a magnetic tape containing the table. With the first type table, this often requires a double
sort -- first into the sequence of the table and second back into the
original sequence. With the second type of table (tabular information),
it usually requires only one sort. The double sort of the first type table
lookup can sometimes be reduced to one by proper planning; the original
sort can place the items in sequence fox the sequential pl'ocessing against
the table . .
Sometimes a third possibility exists; this is to calculate the table values
as they are required. Obviously, this can only be done in those Cases in
which a definite formula can be used to calculate a particular value.
In table lookup operations, the determination of which is faster and
cheaper -- drum reference or tape processing - . . revolves principally around
the time required for sorting. This, in turn, is a function of the number
of words in ,an item and their total numbex. For fairly small volumes and
short items, the additional pre-sorting (if necessary), sequential processing, and final sorting may take less time, ovexall, than the drum lookup
and processing. If the items involved are long, or many in number, the
odds become more in favor of the drum lookup" Standard sorting tables can
be used to assist in computing the relative times, and costs, of the two
methods.
2523.2
Instruction Overflow When programs become too large for the memory
capacity of the equipment, a drum becomes a convenient place to store the
overflow routines, which usually are those to handle more or less '"exception" conditions, or correction sub-routines. These are xead in fxom the
drum as required in a manner quite similax to that cus,tomary when using
magnetic tape for the overflow.
<)
Because of its ttrandom access" charactexistics, the drum practically always makes it possible to obtain an instruction sequence in less time than
is required with tapes. The time saved is largely a function of the number of overflow instruction blocks incorporated into a pxogram.
Original 5/15/57
2523.3
2523.3
Data Overfl.9l!. Occasionally, it is necessary to provide temporaxy stoxage
for part of an item which is too long to contain in its entixety within
memory. For example, infoxmation in the master poxtion at the beginning
of an item may require change based on an action in a txailer poxtion
which may be several blocks away. In this case, the first part of the
item needs to be placed temporarily on tape or dxum until the detail processing has been completed. Either one can be used for this purpose, and
it is standard pxactice in Univac installations to use a "scxatch tape"
fox this purpose.
Comparing Univac II with the 705, the actual time required to write out
the overflow is about the same; drum latency average time and tape start
time in the two systems are about the same, and successive characters are
transferred every 40 microseconds in both systems. However, Univac can
continue reading in and pI'ocessing subsequent blocks during most of the
tape writing period; the 705 can read into a tape control buffer (this assumes it has Type 777 Tape Control Units), but can do no pI'ocessing during
the memoxy to drum transfer. To offset this, Univac would need time to
reWind the overfl,ow tape after the last temporary block had been placed on
it; the 705 does not require a corresponding action with the drum .. Because
of tape reversal times, the 705 to drum method will usually be somewhat
fasteI', but not significantly so ..
2523 .. 4
Summary Comments on Drum Evaluation.. The fact that the 705 can have one
or more drums can be disregarded when building up the machine runs needed
by Univac II to do the job of a potential customex; the goal should be, as
always, the best Univac pI'ocedureo A drum is NEVER A NECESSITY -- it is a
convenience, at times, and can be justified only to the extent that it
pays its own way.. It frequently does save time, and if a potential customeremphasizes processing time without regaI'd to cost, thexe isn't much
that we can do.. Fortunat:e1y, most customers are cost conscious,and they
can be impressed by the fact that each drum must give them an average of
AN EXfRAHALF DAY OF WORK EVERY WEEKo Some savings on one or two jobs are
not enough -- it must justify its cost over a·full base period of operation
as a minimum ..
It has been noted in some instances that an unjustified assumption often
is associated with use of a drum. The fact that it may save time in one
or two I'uns involving table lookup becomes the basis for jumping to the
conclusion that savings accrue whenever it is used. Thus, for example, it
may be assumed that using the drum for overflow program instructions is
the most economical processing method; the fact may well be that extraction
of a work tape and subsequent processing in another machine run is the
more economical approach. One of the pitfalls of efficient EDPM proces~ing
is to view the drum with the same outlook as the mountain climber contemplating Mount Everest -- simply because it's there ..
Original 5/15/57
2529
2529
~ES
POINTS ON THE IBM TYPE 734 MAGNETIC DRUM
These points will be useful in combating IBM proposals involving one
or more magnetic drums in a 702 or 705 EDPM system.
2529.01 A Drum Is a Necessity. The claim may be made that the expansion of memory
because of a drum makes possible a reduction in machine operating time,
facilitates securing overflow instructions and permits "table lookup" type
processes.
For the 734: These are all possible uses of a drum.
Against the 734: See 2523 for the economic considerations which must be
taken into account. A drum is economically justified only insofar as
it "pays its way."
2529.02 Justifying the Cost of a Drum. The monthly rental of a drum is $2,800,
or 10-1~ of the cost of a typical 705 central processing unit.
Against the 734: Restricting consideration to the 705, a drum must add
at least as much productive work as it costs; this is equivalent to a
minimum of another half-day a week. It cannot be justified, on cost
grounds, unless it does add this much output to the installation.
2529.03 Univac II Without Drums Against the 705 With Drums. The best counterproposal to a 705-with-drums proposal isa sound, well-thought-out Univac
method. There is ALWAYS AN ALTERNATIVE MEfHOD OF PROCESSING, and the
development of the Univac procedure need not consider what IBM may 'propose
with drums. The customer's decision will practically never be based on
the availability of drums alone; he is looking for the best data-processing
system. A drum is only one part of the overall installation.
2529.04 Characteristics of the 705 With Drums. These are covered in 2522. In
discussing the counter-proposals with a customer, the~vantages and
limitations of the Type 734 can be outlined. IBM may not always point
out that there is a good deal of lost time in using it with the 705.
Original
/15/57
3110
3110
IBM TYPE 711 CARD READER
The IBM Type 711 Card Reader is designed to read standard 80-column punch
cards in on-line operation only in the IBM Type 701, 704 and 709 EDPMs.
It is not capable of independent off-line operation, but operates always
strictly under stored-program control. Two models, differing in card
reading speed., are available; in general, Modell is used in the 701 system, but can be used with the 704; Model 2 is normally used with the 704
and, so far as available information indicates, is the only version which
can be connected to the 709.
Model 1
Model 2
311001
150 cards per minute
250 cards per minute
Cost.
Purchase
Model 1
Model 2
3111
Monthly Rental
$ 200
$ 52,000
800
CHARACTERISTICS OF THE TYPE 711
With the exceptions noted, characteristics of the two models are identical.
Type of cards read:
Number of columns readable:
Type of card punching read:
Card reading speed (continuous)
Control panel:
Number of reading stations:
Checking of card reading
Memory in card reader:
Method of reading:
Types of selector pickups:
80-column
72
Conventional punch card code or
binary (see 3112)
Model 1 -- 150/minute
Model 2 -- 250/minute
Yes (see 3112)
Two (one for control punches only,
one for reading).
None
No
Card image: 36 columns of one card
row are transferred as a 36-bit
word
(1) Immediate
(2) 11-12 punch ("X" pickup)
(3) 9-12 punch (digit selection)
(4) Card cycle
(5) Split column control
The card feed is a slight modification of that used in the Type 402
Tabulator. The control and reading stations correspond to the second and
third reading stations of the 402.
Original 4/5/57
3112
3112
MElHOD OF OPERATION OF mE TYPE 711 CARD READER
Both models of the Type 711 read 72 columns of an 80-column card and
transmit the information into the EDPM memory as 24 words, each of 36
bits; each word consists of binary IVs or O's corresponding to the presence or absence of a punch in each of 36 columns of one card row. The
memory representation thus is a complete image of the 72 card columns
read.
Through control panel wiring, it is possible to select any 72 card columns
desired qnd to rearrange them in any manner desired to create the 36-bit
words. Variation of word formats (i.e., card columns which are read into
specific bit positions) is possible both within a single card and between
cards through use of the pickup selectors. For the purpose of such data
rearrangement, the control panel is provided with 12 five-position coselectors and two ten-position pilot selectors. To simplify the explanation following, it is assumed that the first 72 card columns are being
read and that the first 36 are converted into one set of 12 words and
the last 36 into the other seto
Each row of a card is read in two separate reading cycles of 36 columns
each; these are called the nleft-read" and "right-read." Specifically,
all 80 columns are actually read in each cycle, but only the 36 columns
wired int a the "entry left" and "entry right" hubs, respectively, generate
impulses which can be transferred to the EDPMo Reading of a card is
9-edge first, with the 24 cycles following each other as indicated in
this illustration:
1
12
11
00
11
2
3
4
5
6
7
8
9
36 37
'23
21
19
17
15
13
11
9
7
24
22
20
18
16
14
12
10
8
5
6
3
4
1
2
72 73 80
Unused
Columns
Once the feeding of a card into the reader has been initiated, the movement of a card through the read brush station is automatic, the successsive 36-bit words becoming available at specified time intervalso The
essential difference between the Modell (150 cards/minute) and the Model
2 (250 cards/minute) is the time interval between successive card rows
Original 4/5/57
3112
(2)
3112
(Continued)
and also between the end of one card and the beginning of the next. The
timing chart for the Model 1 card reader is depicted on the next page;
that for the Model 2 is similar, except for shorter time intervals.
The cards read may be punched either in pure binary or conventional card
codes. In the former case, relatively common in scientific applications,
it is customery to prepare a card with up to 24 binary words, each of 36
bits (sign plus 35 data bits), each word occupying 36 successive columns
of a row. (Normally, columns 1-36 and 37-72 are used). A punch is converted to a binary "1" and no punch to a binary "0". Cards prepared in
this fashion result in a computer word which can be used without change
in computation.
If a card is punched in conventional card codes, the resulting transfer
of 24 words into memory results only in a "card image," and the conversion
program is necessary to transform the information into a form the EDPM
can use. For example, a "6" in card column 36 is transferred into memory
as a binary ttl" in the right-most bit position 6f the 7th word read in;
the program must recognize this particular word as representing the
decimal digit "6" and provide for converting each bit in the word to
the binary representation for a "6. ft The actual program can become very
detailed, since it is also necessary to know whether a particular bit
position represents the units, tens, hundreds, etcl!t, position of a decimal word.
Alphabetic information in punch card code is converted into two binary
"1'5" in two different words transferred into memory" An "E" in column
36, for example, becomes a "1" in the right-hand position of the 9th
word read in (5-row left) and 23rd word (12-row left)" The conversion
of alphanumeric information into a form suitable for processing in an EDPM
basically binary internally requires extensive programming and a considerable amount of computer time. In the Type 701, for example, the conversion of a 72 column input involving some non-numeric fields cannot be
accomplished within the 400 ms card reading cycle.
In either type of card recording, the impulses representing each word are
not stored, but are available only for a limited period of time during
each of the 24 reading cycles on wires reading directly into the EDPMo
If it is not ready to receive the word at that time, the information is
lost. The timing impulses determining when the reading of each successive word occurs are a function of the clock mechanism of the card reader
itself, rather than the EDPMo Because there is a difference in the manner of transferring information from the card reader into the 701/704 as
compared with the709, the two methods are discussed separately in the;
next two sections. The remarks of this section are common to all three
EDPMs o
Original 4/5/57
3112.1
3112.1
Card Reader to Memory Transfers -- Types 701 and 704. In both of these
equipments, the transfer of information from the card into memory requires stored-program instructions~ Two separate ones are involved:
(1) A "Prepare to Read," with an address of the card reader, causes
a card to feed through the reading stations.
(2) A separate ",Copy" instruction for each word of the card to be
transferred into memory. The address of this instruction designnates the memory location into which the word is to be read.
Once the card feed has been initiated, the movement of the card through
the reading station is automatic, and the successive words become available at specified time intervals. The program must provide that "Copy"
instructions are ready for execution at these required times in order to
assure transfer of all information in the cards. Consequently, care is
required to assure that the time limits between the card reader instructions are observed.
A brief discussion of the time limits applicable to the Type 711 Modell
will show the considerations involved. If continuous card motion is
maintained, the complete card cycle will require 400 ms (at 150 cards/
minute). Of this, 292 ms are involved in the time limits the programmer
must observe; the remaining 108 ms are required for execution of "Copy"
instructions and for the necessary tolerances in the mechanical card
feeding mechanisffio
The essential portions of the timing chart, beginning with the transfer
of the last word of the previous card, are depicted:
12 RC
9LC 9RC
8LC 8RC
7LC 7RC
11 LC
12LC
11 RC 12 RC
READ
70 ms
120 ms
These times add up to about 292 ms, the maximum safe available computing
time during the card reading cycle. The various time intervals which are
involved are described briefly.
Execution of "Read" Instruction. For continuous card movement, a '"read"
instruction must be executed within 70 ms after reading the last word
(12 RC =12 right copy) of the previous card. However, the "read" cannot
be executed until 20 ms has elapsed after the 12 RC; consequently, full
card speed with maximum internal computing time is attained if the next
"read" is executed between 20 and 70 ms after the previous 12 RC. If
it is reached in the program before the 20 ms has elapsed, the EDPM
simply "hangs up" until the instruction can be executed; there is in
this case no loss of card speed, but some loss of computing time.
Tolerance Time Before First Word From Card. At full card speed, an average time of 120 ms elapses between reading the 12-row of one card and
Original 4/5/57
3112.1
(2)
3112.1
(Continued)
the 9-row of the next. However, only 70 ms is available for useful computation; the remainder is necessary for tolerances in the card feeding
mechanism. If the card reader is stopped at the time a "Read" instruction is executed, an average of 270 ms elapses before the 9-row is under
the read brushes, but again only 50 ms can be used for computing.
Time Between Left and Right Words of One Row. Each word is transferred
in parallel (36 bits at once) from the read brushes to the MQ-register
and thence into memory by attCopy" instruction. The safe computing time
available between the two words of each card row is 540 microseconds.
Normally, this time is used to modify the memory address of the "Copy"
instruction in the so-called "copy loop. 'f
Time Between Successive Card Rows. 15 ms of useful computing time is
available between reading the right-hand word of one row and the lefthand word of the next row. The nature of the computation is immaterial.
Because the word transfer from the card reader into memory is through the
MQ-register, it is normally necessary to be certain no useful data is
stored there at the time a "Copy" is given; its previous contents are destroyedo The MQ-register can be used, without restriction, during the
computing time available between the successive "copy" instructions.
In the event a word is ready for transfer before a "copy" instruction has
been reached in the program, the EDPM stops with a "Copy Check" light
on the OperatorVs Console. The card continues feeding to the end of the
normal cycleo
Reading of a card may be stopped at any point by executing another "Read"
instruction; the card moves to the end of the cycle, but no additional
words can be transferred into memoryo In this case, the "Copy Check"
light does not go on and the EDPM continues in operationo
If a 25th 1'Copytf instruction
the fact that the end of the
skips to the third following
used as a means of exi t from
counting the number of words
is reached in a program~ the EDPM recognizes
card has been reached, and automatically
instructiono This feature is conventionally
the t'copy loop;" it eliminates the need for
which have been moved into memory.
The various time intervals available for useful computation during the
card reading cycle are beyond the control of the programmero Necessary
impulses are generated by the card reader itself~ and "copy" (or "read")
instructions must be waiting execution when a word is ready for transfero
The limits indicated are the maximum safe computing times available; the
actual times are somewhat greater, but the difference is needen to provide
a sufficient margin of tolerance for the mechanically moving card reader.
It will be noted that a card rate of l50/minute is achieved only in continuous card movement; this requires the next .tread" instruction to be
executed between 20 and 70 ms after the last "copy" for the previous card.
Intermtttent card movement reduces the rate to about 125 cards/minute.
Original 4/5/57
3112.2
3112.2
Card Reader to Memory Transfers -- Type 709. One Type 711 Model 2 Card
Reader can be connected to one channel of each Type 766 Data Synchronizer
in a 709 installation. Thus a maximum of three can be connected to the
system. In this use, it is necessary to have a Type 716 Printer also attached to each channel using a Type 711.
The basic method of card reading is identical to that of the 701/704
EDPMs ~ wi th the exception that the "copy loops·' of those systems are
not necessary; the control circuitry of the Type 766 takes over these
functions. (Refer to 1160 for a discussion of the characteristics and
operation of the Type 766). The successive "words" from the card are
transferred into the synchronizing register of the Type 766, and thence
into memory, using memory addresses and word counts contained in the
Control Word. The essential method of transfer is the same as that applicable to tape units, which is discussed in detail in 1160. As is the
case in tape reading, the transfer of the word from the synchronizing
register into memory is accomplished by "interrupting" the main program
for one cycle; this is completely automatic and requires no attention by
the programmer. The transfer of the word from the reading brushes to
the synchronizing register is controlled by timing impulses generated by
the card reader itself.
The Data Synchronizer Channel can be used for only one input or output
operation at a time. Therefore, when reading a card, it is tied up for
250 ms (at 250 cards/minute), and is unavailable for other inputs or outputs during that period" The card reading may, of course, be interrupted
at any point by one of the means mentioned in 1160, but the card completes
its movement ot the end of the cycle.
3112.3
Checking of Card ReadingQ Both models of the Type 711 have only one
effective reading station and the transfer of information is unchecked.
Similarly, the sensing of control punches in the first reading station
is unchecked. It should be noted that there are at all times (except the
end of a deck of cards) two cards in the feed path; the "read" instruction advances the card which has just passed the first (control punch)
station through the reading station and makes its information available
for transfer to the EDPM. Simultaneously, a card is fed from the hopper
and passes through the first station, where it is read for control punches
only.
311204
Timing for the Type 711 Model 2. A timing chart is not available for the
Model 2 Card Reade~. It is, of course, similar to that of the Modell,
section 311201-, except that most intervals will be shorter. Knowledge of
maximum safe computing times at various points of the card reading cycle
is necessary with the Type 704, which requires, "copy loops," but is not
needed with the Type 709.
, Original 4/5/57
3113
3113
USE OF THE TYPE 711 CARD READER IN BUSINESS DATA PROCESSING
Until the Type 709 is available for delivery (announced for late 1958),
use of the Type 711 Card Readers is limited to the 701 and 704, both of
which are used primarily for scientific computations. There are, however, some 704 installations which are planning to do some commercial
data processing, particularly payrolls; in fact, suggestions from these
users are primarily responsible for many of the internal changes planned
for incorporation in the 7090 The discussion of this section is limited
largely to use of the 711 in a 709 business data processing installation.
Of course, it should be noted that some potential 709 users will be combining both scientific and commercial problems on the equipment. It is
fairly common to find scientific source data in binary form and, in such
instances, uses of the 711 is almost mandatory. One card is capable of
containing up to 24 36-bit words of binary information, and this can be
transferred into the EDPM only through the 7110 Independent card-to-tape
conversion is impossible; all of them handle character codes only, not
straight binary information. If a 709 is to be used partially for scientific calculations, and also if sufficient binary input is involved to
justify a directly-connected card reader, then the 711 may be warranted.
It should be observed that its use demands a Type 716 printer also, and
thus the economics of a 711 is closely wrapped up with a 716; some discussion of the printer is contained in 4110. Because a combined 711/716
costs $2,000 monthly, their total use should justify this expenditure of
money
0
In a business installation (either exclusively or in part), the use of a
Type 711 must be weighed against an off-line Type 714 Card Reader, which
with a tape unit aggregates a minimum of $2,950 monthly. This section
discusses some of the factors to be consideredo
311301 Effect of Alphanumeric Nature of Business Datao Business data are almost exclusively decimal or alphabetic in nature; although master files
on magnetic tape may be maintained partially in binary (at least on the
709), basic inputs and outputs consist of decimal digits, letters and
special characters. Most of us, certainly, are not particularly facile
at reading $5,621.87 in binary notation--l000100lOl00000010ll. Inputs
and outputs in commonly-understood alphanumeric representation are mandatory.
Because the Type 711 transfers only a 24-word "card image" into memory,
an extensive conversion program would be necessary to create binary-coded
decimals or, in the case of numeric data, binary words. How long this
will take has not yet been determined; as a point of comparison, the 701
could not convert 72 characters in the 400 IDS card reading cycleo Although
the 709 is faster internally, it is rather evident that it will spend
considerable time, and require a fairly lengthy program, to do this conversion. Effectively, then, it will be tied up for a considerable part
of the card reading cycle on transforming the card image input into a
usable form. By reducing its ability to do other productive work (using
other Data Synchronizer channels for input and output), the 709 becomes
an extremely expensive input deviceo It may be useful to develop a simple
Original 4/5/57
3113.1
3113.1
(Continued)
economic comparison. The question to be answered is this: Which is
cheaper, reading cards directly or through a Card-to-Tape conversion?
Suppose the 709 CPU rents for $50,000 monthly (somewhat on the low side).
Further assume that the internal conversion of a card read on the Type
711 takes 100 ms--which probably is not too far from the actual time.
This means that 40% of the card reading cycle (about 240 ms) is spent on
transforming the input information. Thus, during the time it is reading
cards, the 709 is costing the equivalent of $20,000 a month for this purpose. An off-line C-T converter costs about $3,000 a month--15% as much.
Thus if the 709 spends more than 15% of its time reading cards, the percentage of its costs chargeable to card reading exceeds that of off-line
operation. 15% of an 8....hour day is 1.2 hours, time for 18,000 cards at
full reading speed. That's a fairly modest card volumeo
3113.2
Character Transfer Rate. At 72 characters maximum per card, the transfer
rate of information into memory, at full reading speed, is about 300
characters per second. This is pretty slow. Although it may not be
critical in some applications--for example, posting a low-density file-in others it may cause considerable EDPM "stalling" waiting for the next
card to be read. This is an example of the typical case. of the mismatch
occurring between a slow input/output unit and a much faster CPU. As a
further note, this transfer rate is about 2~ of that obtainable from
tapes. Not only is one Data Synchronizer channel working at much less
than capacity, but the internal computing time required to transform the
card image is many times that necessary for a tape resulting from C-T
conversion.
3113.3
Lack of Checking. A major disadvantage of the Type 711 in a business
process is its complete lack of checkingo A card is read once-and only
once. Because the reading mechanism is electro-mechanical, the error incidence to be expected is much higher than in internal EDPM processes.
There is no economic method of assuring accurate card reading, or of detecting errors when they do occur.. Batch proof totals may be used to
check the accuracy of reading the fields involved, but there is no certainty that the other (an usually more numerous) columns have been correctly read.. Foolproof error-detection could be provided only by accumulating "hash totals·' for all card fields, recreating these totals
internally as the card; are read, and comparing against periodic "hash
total" cardso Such a procedure would indicate an error, but would not
provide for correcting it. It also is uneconomic; the proof totals would
have to be computed prior to the card reading, on equipment external to
the EDPM; at least one special card pass would be necessary on a tabulator
to compute these totals, and more might be required.
Wi thout such "hash totals,'· it is extremely doubtful if any auditor would
accept any accounting procedure using Type 711 input to an EDPM. As a
useful means of data input, this unchecked card reader is definitely unattractive.
Origina14/5/57
3119
3119
SALES POINTS ON THE TYPE 711 CARD READER
Index numbers 3119001--3119049 are reserved for pointers applicable
largely to business operations; 3119050--3119099, for those of use
primarily in scientific applicationso
3119.01
Lack of Checkingo The Type 711 (both models) has only one effective data
reading station. There is no verification that the word images are a
correct representation of the actual punches in the card.
Against the 711; In business operations, particularly applications in
which accuracy is mandatory or highly desirable, extensive external
checking features must be incorporated into the processing system to
assure correct card readingo Any feasible method of providing such
checks can be only an error-detection scheme, which would not indicate the specific card read incorrectly, but only isolate an error
to a batch of cards for which a series of proof or "hash" totals has
been generated prior to the card readingo
3119002
Reading of Alphanumeric Informationo The Type 711 reads a card and
transfers the data into memory as a series of 24 "card images," each
representing 36 columns of one card row.. Information in standard punch
card code requires programmed conversion into either internal character
codes or pure binary numbers ..
For the 711; The reading method permits easy handling of pure binary
informationo In addition, since a complete (72-column)card image
is moved into memory, multi-punched columns can be read without
special handling.
Against the 711: The internal program time required to convert the
24-word card image into a usable internal language is considerable.
Business data is almost exclusively in punch card codes, not binary.
3119~03
Elimination of a C-T Converter if a Type 711 is Otherwise Required. In
proposals in which a Type 711 may be necessary to read binary scientific
input, IBM may propose that this reader be used in lieu of a C-T converter (Type 714), simply to save the $2,950 cost (including tape unit)o
For the 711* This is a possible solution 9 and may result in a saving
of about $3,000 in basic installation rentalo
Against the 711: There are two rebuttalso
(1) With the 709, a Type 716 is required with a Data Synchronizer
channel before the Type 711 can be connectedo This boosts the
cost of the two units to $2,000 monthly, rather costly unless the
printer can be utilized economicallyo
(2) The proposal is economic only for small card volumes, because of
the extremely high CPU cost of converting alphanumeric card input.
For any specific proposal, an analysis similar to that of 3113.1
can be developedo
3119,,04 Mismatch of Card Reading Rate and Computing Speedo The usual remarks on
the speed mismatch between the relatively slow card reader and the fast
CPU are applicableo Because computation time and card rate seldom are
equal, one or the other is used at less than maximum efficiencyo
Original 4/5/57
3130
3130
IBM TYPES 712 AND 714 PUNCH CARD READERS (CARD-TO-TAPE CONVERTERS)
The IBM Types 712 and 714 Punch Card Re'aders are designed to read standard 80-column punch cards and transfer the reading, in 702/705 character
codes, either into 702 or 705 memory, under program control, or onto magnetic tape in an off line card-to-tape (C-T) conversion. The resulting
tapes may be used as input to a 702 or 705, as well as to a 650 or 704,
provided tape format restrictions of the latter two EDPMs are met.
The Type 712 is an early version provided with the first Type 702s delivered. It has been superseded by the Type 714 and, as far as known, is
no longer being manufactured, although some are still in use. It differs
from the 714 in not having a plugboard and, as a C-T converter, in not
verifying the accuracy of the tape writing. The discussion of this section is limited to the 714.
Both units consist of two physical cabinets: A card reader and a control
unit providing power, timing and decoding circuitryo A Control Unit is
reqUired for each Type 712 or 714.
3130.1
Monthly Rental:
Type 712
Type 714
$1,050 (including Type 756 Control Unit)
2,400 (including Type 759 Control Unit)
There is a current tendency on IBM's part to quote separate rentals for
each type number of an associated and integral group of type numbers. The
712 is separately quoted at $750 monthly and the 756 at $300, totalling
the $1,050 given above.. Separate figures are not available for the
714/759; however, both must be used together.
When used in an off-line C-T conversion, both require a Type 727 Tape
Unit at $550 monthly additional rental.
3131
BASIC CHARACTERISTICS OF THE TYPE 714/759 CARD READER
Purpose of equipment
Type of punch card
Type of buffer storage
Character capacity of buffer
Conversion of punch card codes
Card reading speed
Magnetic tape unit (off-line)
Transfer rate, buffer to tape
Transfer rate, buffer to memory
Plugboard editing features
Multi-card conversion
Checking of operations
Length of converted record
Direct on-line card reader or off-line
C-T conversion
IBM 80-column cards only
Magnetic core
92 characters
Automatic, to 702/705 character codes
250 cards per minute
Type 727
00067 ms per character, plus start
0.0335 ms per character (702 and 705)
Yes; see 3132
One or two cards, not to exceed 92
characters total
Yes -- all are checked
Variable, up to 92 characters
In an off-line conversion, all normal characteristics of the Type 727
Tape Units are unaltered (see 1101).
3132
3132
METHOD OF OPERATION OF THE TYPE 714/759 CARD READER
This section describes the plugboard editing features and briefly discusses
on-line and off-line operation.
3132.1
Type 714 Control Panel. The entire operation of reading a card into the
714 buffer is controlled by plugboard wiring.
1.
Reading Stations. There are two reading stations. In the first, a
series of check bits are created. The second reading is the one actually transferred into,,·,the buffer after comparing with the check
bits for accuracy. Wiring from the reading brushes into the storage
entry hubs is completely flexible, and any desired rearrangement of
the card format can be accomplished.
2.
Column Splits. Twelve sets of column split hubs permit separating
"11ft and "12" control punches from 0-9 information punches, and either
recording them in separate character positions of the buffer, or
eliminating them.
3.
Digit Emitters. Four digit emitters are provided: Two are available
on the "12 f1 cycle only, one on the "11" cycle, and one on the 0-9
cycles. These automatically generate impulses on the applicable card
row cycles and permit including in the buffer record characters not
punched into the card. The 0-9 emitter is normally wired through a
digit selector to select the specific digits wanted.
4.
Digit Selectors. Two digit selectDts are available. These permit
breaking up a multiple-punched card column into several different
character positions in the buffer, and are also used in conjuction
with the 0-9 digit emitter to create constant characters not present
in the cards.
5.
Record and Storage Mark Emitters. These are special-purpose emitters
for creating the punch-card pattern for a record mark (which may be
in the card also) and a record storage mark (RSM), a character code
pecul~ar to the Type 7140
6.
a.
In buffer-to-memory or buffer-to-tape transfers, the operation is
terminated by the RSM. Thus only the number of characters required is written out, provided the RSM follows the last data
character in the buffero If there is no RSM, the entire 92 characters in the buffer are transferred. The RSM is not placed in
memory or on tape, but simply terminates the transfer operation.
(The character code of the RSM is not known, and is immaterial.
Its use is confined to the Type 714).
b.
Emitting of record marks is provided solely as a convenience in
future EDPM processing. They are handled as any other valid
character code in buffer-to-memory/tape transfers.
Alternators. These are used when two caxds are to be converted into
the buffer before its contents are transferred into memory Or 6nto
tape. Up to 46 characters from each of two successive cards can be
3132 1
0
3132.1
(Continued)
handled in this manner. Identical columns must be read from each of
the two cards in exactly the same manner; the alternator is essentially
a 46-position "selector'" which automatically shifts from one state to
the other on each card cycle, enabling the two successive (partial)
card images to be converted into different buffer positions.
All control panel function hubs are duplicated to permit verification of
the card reading in the two read stations; this includes the emitters.
(Instruction manuals state that there are 24 column splits and four digit
selectors. However, because checking of the second reading against the
first is automatic and includes everything placed in the buffer, these
two features always require use of hubs in pairs 1 reducing the effective
number to 12 and 2, respectively).
3132 .. 2 Checking of Card Reading.. Results of card reading at the two stations
are checked adequately. The method of checking differs appreciably from
that of the Univac C-T converter, but is considered fully satisfactory.
3132.3
Checking of Buffer-to-Memory/Tape Transfers. Direct transfers from the
buffer into memory receive a check bit test only. Because the buffer
record has already been verified, only a double error in the transfer of
a single character code would be undetected~ The probability of suchan
error is too small to worry about~
Transfers from buffer to tape, in C-T conversion, likewise are checked,
but again in a manner different from Univac.. The tape recording is reread for both check bit and check character agreement, but is not compared against the actual buffer record.. Again, the probability of an undetected error is negligibleo
3132.4 Card Capacity per Tape Reel. The card capacity of a magnetic tape reel,
in C-T conversion, varies with the length of the records converted and
whether one or two cards are placed in each0 These are typical maximum
capacities for a full 2400-foot reel of tape; normal capacity is about
5% less ..
Cards per Record
Two
Characters Qer Tape Record
One
40
60
80
92 (maximum)
30~300
27,400
25,000
23,800
60,600
54,800
50,100
47,600
For two cards per tape record, the number of characters from each card
is half the figure in the left-hand column.
3132.5
Direct EDPM Operation of the Card Reader.. The card reader is capable of
operation hooked directly to either a 702 or 705 (with a 759 Control Unit)
under stored-program control.. In this type of use, the transfer of the
contents of the buffer into memory automatically initiates the action of
reading the next card (or cards); plugboard features are all operative
during the conversion into the buffer.. Up to 100 card readers can be
handled simultaneously.. Transfer of the buffer record into memory
3132.5
3132.5
(Continued)
is accomplished in the conventional manner: A "Select" instruction, with
an address in the range 0100-0199, alerts the specific card reader desired;
a following "Read" instruction starts the transfer. This is executed at
the rate of 0.0335 ms per character (in both 702 and 705), plus instruction time. This is approximately 30 characters per millisecond, rather
slow compared with the internal speed, and as usual no other operations
can occur during this transfer time.
An error in either the card reading or transfer from buffer to memory
turns on the tt0902 Read-Write Check Indicator, tt. which can be set to automatic stop or to program control~ The indicator does not distinguish between the two sources of error; therefore it is usually necessary to feed
the card just ejected into the stacker as well as the two at the read
stations back through the card reader.
Although the card reading rate of 250/minute (240 ms per card) is extremely slow compared with the EDPM speed, IBM has rather consistently
advocated a directly-connected reader for loading machine programs, and
for certain types of data input, such as those in which the time to find
master file items stored on magnetic tape averages more than the card
reading time (low activity file maintenance operations are typical). This
latter use is probably secondary to the loading of program cards; if a
714 is to be connected for this purpose, it might as well be used for
something during the main body of a machine run. The justification for
maintaining programs in punch card, rather than magnetic tape, form is
that corrections can be key-punched and manually inserted, without tying
up the central processor; if programs are stored on an instruction tape,
an EDPM run is obviously necessary to update or correct them.
3132.6
a Card-to-Ta e Converter. As an off-line operation, the
used with a standard Type 727 Tape Unit as a card-to-tape
converter. There are no particularly significant features in this mode;
it operates basically the same as the Univac C-TConverter. The tape
prepared in this operation is acceptable to either the 702 or 705, as well
as to the Type 650 and 7040 For the latter two equipments, the conversion
must be into a tape format acceptable to the respective EDPM (see 1103).
There are, at this time, no known plans to use it with the 650, which
has other (and cheaper) card readers in its periphery array. Its cost
is probably too high to justify including it in a 650 installation unless
special considerations make independent card-to-tape conversion necessary.
There is, however, no technical reason to prevent its being so used.
3175
3175
IBM TYPE 537 CARD READ-PUNCH UNIT
The Type 537 Card Read-Punch Unit is an auxiliary input-output device for
the Type 650 EDPM, which is designed to permit both reading from and
punching into a single card. It thus differs from the Type 533 ReadPunch Unit, which cannot punch into the card being read.
3175 1 Monthly Rental:
<)
$ 7000
The 537 requires one input-output synchronizer on the Type 650 drum; one
of these is included in the basic 650 rental, with two more optional
at $300 monthly each. Normally, the 537 will require an additional
synchronizer
Alphabetic character devices may also be extra; see 6113.
0
3176
BASIC CHARACTERISTICS OF THE TYPE 537
The 537 card feed is essentially the same as the card-feed units used
with IBM's Types 604, 607 and 608 calculating punches. Functional operations, however, are quite similar to those of the 533.
Card feeds
Card speed (maximum)
Number of card stations
Type of control
Number of stackers
Handling of error cards
One
155 per minute, read and punch
Five
Plugboard and 650 stored program
One
Offset in stacker
The plugboard (control panel) layout and terminology is identical with
the one used with the Type 533 (see 3172), with the exception of a pair
of t~swi tch'" hubs wired if the 537 is used for punching only. Wiring
principles are identical with those of the 533, with a few exceptions.
In principle, the first and second reading stations of the 537 are
treated in about the same manner as the read feed of the 533; the punch
and punch read stations correspond closely to the punch feed of the 533.
3177
3177
METHOD OF OPERATION OF THE TYPE 537
The basic use of the Type 537 follows closely the principles of the Type
533. The same plugboard is used, although a few hubs have different
interpretations
0
3177.1
T~Qe
537 Card Feed.
rOWE~1
The card feed is graphically portrayed thus:
1:, R<£"
ReAOING7
./
(if I
5
PtJNCf/
SECOND
Re;.A-D/NG
\\\
j
.(-#=3
#4-
~ "\
CALC.
J
--
PUNcH
~
ReADING
\(#
TACkT
1.
650
gENERAL.
OV-rPVT
SYNCH.
There are five cards, one
loading and final runout);
by feeding ina .card from
"Write" instruction. The
~- SrORAGrE
at each station, at all times (except initial
advance is initiated for all five simultaneously
the hopper. This is done by a 650 ttRead tt or
five card 'stations and their uses are:
Card #1 is checked for double punches and blank columns in the Punch
Reading station. Information in this card may also be plugboardwixed to gang-punch into the card at the Punch Station.
Card #2 is punched from the output synchronizer as it passes through
the Punch Station. This normally is the result of computation performed by the 650 during the preceding caxd cycleo
Card #3 is in a tfwai ttt. status at the Calculating Station; . the 650 is
computing on data pertaining to this card during this cycle.
Card #4 is being read into the 650 input synchronizer in the Second
Read station.
Ca~ is being read for control punches and selectoT pickups in the
First Read station.
The control panel may be wired to read a card into the 650 input synchronizer from either the first or second reading station. The latter is
normally used.
The 537 may be used for reading only, for punching only, Or for both
reading and punching.
4130
4130
IBM HIGH-SPEED PRINTERS -- TYPES 719, 720 AND 730
The IBM line of high-speed printers designed for EDPM use consists of
three similar versions differing only in maximum printing speed and
number of characters per line:
Type 719
1,000 lines per minute, 60 printing positions
~
tt
Type 720 -- 500·
"
«
, 120
ff,
Type 730 -- 1,000»
»
»
, 120"
The basic method of operation is identical for each type and all are
discussed in this section. Each can be connected directly to a Type 702
or 705 EDPM, in program-controlled operation, or used in off-line printing in the same manner as the Univac High Speed Printer. In either case,
each HSP requires one Type 760 Control and Storage Unit to provide powex,
control functions and storage facilities; off-line use requires also a
Type 727 Magnetic Tape Unit, but two may be used. None of the printers
can be used at any time without the Type 760.
In off-line operations, output tapes from the Types 650 or 704 EDPMs
can also be used as pxinter input, subject to certain restrictions which
are discussed subsequently.
4130.1
Monthly Rental:
Type 719
Type 720
Type 730
$ 1,400
1,400
2,100
Total monthly rental for working printer assemblies, including the control and tape units, are:
On-Line
Operation
Types 719/720
Type 730
Off-Line Operation
1 Tape Unit
2 Tape Units
$ 3,200
$ 3,750
$ 4,300
3,900
4,450
5,000
These costs are for each printer. The figures for on-line operation
do not include the costs of any tape units which may be associated with
the 760 controlling the printer.
4131
4131
PRINTER CHARACTERISTICS AND METHOD OF PRINTING
The characteristics and features in this section are common to all three
printers, except for differences noted.
4131.1
Basic Printer Characteristics and Common Features
Maximum printing speed
Number of printing positions
Characters printed
Printing positions per inch
Vertical line spacing
Tape-controlled carriage
Type of paper feed
Maximum form dimensions
·Mlnimum form dimensions
Paper skipping or fast-feed
Number of carbon copies
Storage of information to be
printed
Plugboard editing features
Character suppression
Checking of printer accuracy
1,000 lines/minute (719 and 730)
500 lines/minute (720)
120 (720 and 730)
(60 (719)
47 (figures 0-9, letters A-Z, special
characters & • - $ * / , % # @ l:1 )
10 (6-inch print line on 719, 12-inch
. maximum on 720 and 730)
6 per inch
Yes; three form spacing options:
(1) Single space (6 lines/inch)
(2) Double space (3 lines!:inch)
(3) Tape control
Continuous form, standard marginal
punching pin-feed
Width - 16 3/4" overall, including
marginal strip
Length- 22 inches maximum
Width - not stated
Length- no minimum
Yes, but may cause drop in printing
speed
Original plus seven
Stored in Type 760 buffer
None; no plugboard
Print suppression levers permit
supressing specific groups of
print positions
Checked for accuracy of character code
transfers and character printed
The discussion which follows requires an understanding of the method of
using the s.torage buffer in the Type 760 Control and Storage Unit; this
is covered in 1150.
4131.2
(1)
4131.2 Method of Printing. Printing is accomplished by creating each character
pattern from a print matrix or printing head of 35 wires, arranged in a
5x7 grid~ Individual characters are formed by extending a matrix pattern
of wires which differs for each of the characters permissible. This illustration is a head-on view of a print matrix, with lines connected the
wires extended to form the letter "H:"
• • •
• • •
• • •
• • •
• • •
• • •
MechanicaLLy, the formation of a character is accomplished by running the
matrix wires back into rotating "'code rods; n these have grooves and lands
which cause some wires to project and others to remain retracted. These
code rods are rotated and stopped at the angular position that has the
groove and land combination creating a desired character. The actual
mechanism, needless to say, is a good deal more complicated than this
simple explanation.
I
On the Type 719, there are 30 of these print matrices, placed in alternate
character positions; that is, spaced five to the inch. In one print subcycle, the odd-numbered characters are formed and printed; for the next
'
Yes
No :'
Yes
Yes
Yes
No
100
100
100
30
100
2,400
1,050
1,800
2,800
550
2,000
1,850
2,200
3,000
1,400
1,400
2,100
75
I
Maximum
Number
On-Line
10
i
100
Address
Range
On-Line
0100-0199
0300-0399
0400-0499
1000-9999
0200-0299
0200-0299
0200-0299
0200-0299
0600-0699
02X4
02X4
02X4
0500-0599
Maximum
Number
Tape Units
*
10
2
1
8
*The maximum number of tape units in a 705 installation depends upon which tape control units are used. With ten
Type 754's, 100 tape units are possible; this is the upper limit. There is no limit in the number of units which
may be used off-line.
IBM TYPE 70~ EDPM
MAXI MUM OF 10 TYPE 754,760. '74
a 777 POSSIBi..E, EACH CONTROLLING
NUMBER
OF
7271
SHOWN.
734
60,OOOCHARACTOR
MAGNETIC D~UM
CONTROL
S
745
UNIT
2800
POWER
SUPPLY
UNIT
r---"-'--'"71
i
!,
777
I
CARD
I!
TAPE
READER
100
• : ••• .- .' .' :
MAX.
.. ' . .
754
734
o
:':
... ....
r ..'. .....
.
.
~.
~
.
.
RECORD
.....
j
4
.
..
TAPE
•••
. ,".
CONTROL
...... ' .
r-:', '.~:
. COORDINATOR
~
,
"
"
UNIT
705
CENTRAL PROCESSING UNIT
MODEL I
_~:ru::;;;:!;-;:!$:;:re:SiU.l::;"120'000
CHARACT E RS
40.000 CHARACTE RS
~
760
RECORD
MODEL 2
747 1174
a
STORAGE
.
UNIT
'.~. ~ "
.. :. ',:.'. -
..
..
. .. ..... .
'"
782
CONTROL
UNIT
OPERATOR'S
CONSOLE
100
MAX. PRINTER
519
cARD
PUNCH
100
MAX.
REPRODUC t NG
PUNCH
TABULATOR
31 JAN 1957
5142 .. 1
5142.1
A "normal't Type 705 installation will
t'Normal t' Type 705 Installations
vary considerably in cost, depending not only upon the number of tape
units and types of control units, but also upon the array of periphery
equipments required. For many applications, one card reader at 250 per
minute and one punch at 100 per minute are enough; there may even be
little use of the latter. Similarly~ the nature and volume of printing
will affect the choice of a printer and its cost~ With a print volume
of not more than about 70,000 to 80,000 lines per shift, this example may
be considered a roughly representative installation:
0
1
2
10-16
1
1
1
Type
Type
Type
Type
Type
Type
705
777
727
714
722
717
.
..
.
CPU, 20,000 characters"
" ..
Tape Record Coordinators ..
Tape Unitso .. " " ..
Card Reader with 759 Control.
.
Card Punch with 758 Control
Printer with 757 Controlo ..
.. ..
0
0
.
.
Total monthly rental
0
8
0
•
•
0
.
..
$ 14,000
6,000
5,500 - 8,800
2,400
1,050
1,800
.$30,750 - 34,050
0
Because tape units are necessary to operate the peripheries off-line, ten
is a relatively low complement r- 12 or 13 is probably the minimum number
which can provide efficient CPU and off-line operation. The card punch
may be omitted, but still leaves the cost of a reasonably good installation at more than $30,000 monthly.. When the addition of one or two Type
734 magnetic drums, replacement of the Type 717 by a higher-speed Type
720/760 printer, and possible inclusion of a Type 774 Tape Data Selector
are taken into account, it becomes apparent that $35,000 monthly is a
mOre appropriate rental for an average basic 705 installation.
5142.2
Stripped-Down Type 705 InstallationsG To combat the proposal of a Univac
1 in areas within its potential scope, IBM has proposed greatly strippeddown 705 installations strictly to meet the Univac I cost. Typically,
such an installation may consist of:
1
1
7
1
1
Type
Type
Type
Type
Type
705
754
727
714
717
$ 14,000
CPU, 20,000 characterso
..
" .. ..
2,000
Tape Control Unit
..
3,850
Tape Units
..
..
.. .
..
2,400
Card Reader with Type 759 Control
1,800
Printer with Type 757 Control . " .. ..
0
0
.
0
/>
Total monthly rental ..
0
0
0
..
/>
$ 24,050
This is just about bedrock for a 705 installation.. It provides no buffering of tape input and output (this could be made available for another
$1,000 monthly), which means that all computation occurs outside tape
time. The limited number of tape units necessitates quite inefficient
processing on many jobs; sorting, for example, would be restricted to
a two-way merge, rather than the faster 3- and 4-way merges. There,is
no card punch output whatsoever and only a 150-line printer. A Univac
I of exactly the same units would run $21,485 (which includes buffered
input and output, a 600-line printer and two more tape units for the
printer and card reader)., A full-fledged Univac I with ten on-line Uniservos, a T-C and C-T converter and an HSP would run only $24,.160, or
little more than the stripped-down 705.
Original 2/15/57
5142.3
514203
Maximum Type 705 Installation. To provide its best possible cost-of-_
performance ratio, IBM may propose a wider array of equipments than either
of the preceding two versions. They are particularly common in proposals
hinging on the handling of extremely large master files.
.
1
3
16
1
1
1
1
1
Type
Type
Type
Type
Type
Type
Type
Type
705
777
727
714
760
720
722
734
CPU, 40,000 characters ..
Tape Record Coordinators .. ..
..
Tape Units.
"
Card Reader with Type 759 Control
Record Storage Unit.
Printer
"
"
Card Punch with Type 758 Control.
Magnetic Drum •
•
'" '"
Total monthly rental.
····
.....
...····
. ..
.. ··· ..
..
. · .
0
$ 16,500
9,000
8,800
2,400
1,850
1,400
1,050
2,800
• $ 43,800
The sales emphasis by IBM on a proposal of this general configuration-indeed on any proposal involving two or more TY1pe 777s--:-:is on the faster
tape speed claimed to be possible by paralleling input and output in two
or three of them. This is discussed exhaustive:ly in 1160, which develops
detailed time and cost data for this type of operation, together with'
comparisons against Univac II performance. The use of the third IRC
in theory makes even greater processing speeds possible, but the analysis
of 1160 indicates definitely that the percentage increase is not equivalent
to IBMos claims. A 705 proposal involving three IRCs may indicate that a
job can be done faster than on Univac II, but the cost of doing a unit of
work will be greater.
Subsequent paragraphs in this section will bring out additional points
on the cost-of-performance ratio of the IBM 705. The three "typical"
configurations drawn up are only indicative of the minimum, average and
maximum proposals which may be encountered in meeting IBM competition.
Original 2/15/57
5150
5150
IBM TYPE 709 ELECTRONIC DATA PROCESSING MACHINE
The IBM Type 709 is a large-scale EDPM designed for both business data
handling and scientific computation. It is a further extension of the
Type 704, having the same basic logic and operating speed, but incorporating a number of new instructions to facilitate the handling of businesstype information and providing greatly improved input-output facilities.
First deliveries are scheduled for the latter part of 1958.
The type designation 709 refers specifically to the Analytical and Control Unit, without the high-speed storage (memory) or power supplies
necessary for an operative central processor~ In common usage, it also
refers to an EDPM installation having a 709 as the central processor.
Information available is not specific as to the exact units necessary to
form various configurations of the central processor~ Those known or believed to be associated with the main frame are included in the price
listing of the next section. The use of the Type 736 Power Supply Unit
is uncertain, but it probably is required at all times& The difference
between the two models of the Type 741 also is not known; a "P,ossible explanation is that the more expensive one is used with the 32,768-word
magnetic core storage, and the cheaper with the smaller ones.
5150.1
Cost. Rental and purchase prices of the various units necessary to
make a central processor are:
Purchase
.
$ 10,000
1,100
$ 600,000
Type 709 -- Analytical Control Unit. e
57,200
Type 736 -- Power Supply Unit.
.. "
First 4,096 words of Magnetic
Type 737
208,000
Core Storage (Model 3) .. "
Type 737
Next 4,096 words of Magnetic
193,000
Core Storage (Model 4) .. "
Type 738
32,768 word Magnetic Core
1,040,000
Storage. ..
.
e
72,800
Type 741
Power Supply (Model 2) .
96,000
Type 741
Power Supply (Model 3)
Power Distribution Unit
Type 746
67,600
(Model 2 or 3) "
0
0
.
0
.
0
·
·
·
0
0
Rental
4,000
3,200
20,000
1,400
1,600
0
1,300
Assuming that a Type 736 is necessary in all configurations and that the
Type 741 Model 3 replaces the Model 2 with the large core memory, these
prices result for the three versions of the 709 central processor:
Type 709
Type 709
Type 709
4,096-word core storageo
8,192-word core storage..
32,768-word core storage
0
..
$1,005,600
1,198,600
1,860,800
$ 17,800
21,000
34,000
Original 2/15/57
5151
5151
BASIC CHARACTERISTICS OF THE IPM: TYPE 709
The characteristics included here are limited to those essentially a part
of the central processor. Characteristics of the periphery units are
part of the discussion of each of them, and some of these can be con-"
sidered as main frame attributes. Refer to 5152 for a chart of all
equipment type numbers which can be associated with the 709, either onor off-line, and to 0910.1 for a cross-reference between type numbers and
manual sections.
Magnetic core
4,096 word (Type 737 Model 3 memory)
8,192 word (Type 737 Model 3 and 4)
32,768 word (Type 738)
Word length:
36 bits (fixed)
Information representation:
Binary (35 bits plus sign bit) or
character code (six 6-bit codes)
Memory addressability:
Word-addressable; 15 bits or five octal
digits maximum address
Type of operation mode:
Binary on one word
Type of memory access:
One word in parallel
Type of instruction:
Single address modified to include Index
Register of B-Box address
Arithmetic operation:
Binary
Decimal-to-binary conversion: Table lookup
Binary-to-decimal conversion: Table lookup
None; there are no check-bits attached
Internal checking:
to words
12 microseconds (includes access and
Basic cycle rate:
operation)
Instruction execution times:
24 or 36 microseconds for almost all instructions except multiply and divide
Registers:
(1) Accumulator (p; Q, 1-35)
(2) MQ Register (S, 1-35)
(3) Storage Register (essentially highspeed bus)
(4) 36 Sense Indicators
(5) Instruction Register
(6) 3 Index Registers (B-Boxes)
Direct inputs into memory:
(1) Magnetic tape
(2) Punch card reader
(3) Magnetic Drum
Direct outputs from memory:
(1) Magnetic tape
(2) Magnetic drum
(3) Cathode ray tube
(4) Printer
(5) Card punch
Type of storage:
Storage capacity:
Original 2/15/57
5152 EQUIPMENT CONFIGURATIONS POSSIBLE IN A TYPE 709 INSTALLATION
The chart on the next page depicts graphically the various equipments which may be associated with a Type 709 in on-line
and off-line operation. Certain pertinent data are summarized in the table below.
Type
Number
Description
Monthly
Rental
Operation
OnOff-
Maximum
Number
On-Line
~
~
$4,000
7,200
20,000
Yes
Yes
Yes
No
No
No
1
1
1
Units
800
Card Reader (250/min)
1,200
Printer (150/min)
600
Card Punch (IOO/min)
650
Magnetic Tape Unit
1,800
Tape Control Unit
3,500
Data Synchronizer
Magnetic Drum $2,8000r3,500
2,700
Cathode Ray Tube
150
Cathode Ray Tube
550
Magnetic Tape Unit
2,200
Tape Data Selector
1,850
Record Storage Unit
1,400
Printer
1,400
Printer
Printer
2,775
1,800
Printer
1,050
T-C Converter
2,400
C-T Converter
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
No
No
No
No
No
No
No
No
No
No
No
No
Yes
No
No
No
No
No
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
3
3
3
48
6
3
2
1
1
Address
Range
On-Line*
Maximum
Number
Tape Units
Magnetic Core Storage Units
737-3
737-3&4
738-1
Input-Output
711-2
716-1
721-1
729
755
766
733
740
780
727
774
760
719
720
730
717/757
722/758
714/759
4,096 words
8,192 words
32,768 words
00000-07777
00000-17777
00000-77777
OX321
OX361
OX341
See note**
Non-Addressable
See note***
00301-00310
8 each
16 each
*Octal addresseso "X" is 1, 3, 5.
**Magnetic tape addresses depend upon Data Synchronizer Channel, mode of writing and tape unit number.
***See Section 1160.
5152.1
5152.1
Equipment Configurations in Type 709 Installations~ The specific array
of periphery units and size of memory in a 709 installation can be expected to show rather wide variation, depending upon the major purpose of
the operationo A customer with predominantly scientific uses for the
equipment may sacrifice on some of the input-output devices, because these
requirements are only nominal~ Conversely, an installation devoted mainly
to business data handling may be of considerably different composition.
The possible installations shown in the next few sections are only roughly
representative and will be revised as soon as IBM indicates a pattern in
its sales proposals, and the actual arrays of equipments ordered by the
first few customers are availableo
5152.2 Minimum Type 709 Installation, Scientific Purposes. Desires expressed by
users in the past indicate that the largest feasible internal memory is
a definite attraction in the scientific computer field. With the 32,768word Core storage memory announced for both the Type 704 and Type 709,
the exact place of each equipment in the marketplace has ·not yet been
evaluated. The essential difference is the faster input-output rate of
the 709 (in terms of internal computing time required), although this newer
equipment also has some advantage in internal arithmetic operations because of its more extensive instruction repertoire. The relative speed
improvement in typical mathematical computations is not known.. However,
a minimum-type 709 may be predicated upon this equipment array:
.
Type 709
Analytical Control Unit .. '"
..
Type 736
Power Supply Unit
.. .. " ..
Type 741
Power Supply Unit (Model 2)
Type 746
Power Distribution Unit.
Type 737
8,192-Word Magnetic Core Storage.
Type 733 -- Magnetic Drum ..
-a
.. .. .. ..
Type 766
Data Synchronizer
.. "
" '" .
Type 711
Punch Card Reader (Model 2)
.
Type 716
Printer . .. ..
Type"V21
Card Punch" " ..
Type 729
Magnetic Tape Units " "
"
Type 755
Tape Control Unit" .,
"
.. ..
$ 10,000
8
.
01
~
II
II
.
"
....
..
. " ..
0
0
0
8
0
.. 'iII·o
Total Monthly RentalQ
0
.. .. ..
o
1,100
1,400
1,300
7,200
2,800
3,500
800
1,200
600
"5,400
1,800
$ 37,100
This probably is about bedrock for a 709 installation.
5152.3
Minimum Type 709 Installation, Business Purposeso The approximate minimum
709 installation for business data processing is predicated upon the fact
that sufficient input-output units must be provided to justify the cost
of the configuration. The theoretical internal speed of the 709 is quite
fast, and the new instructions permit a rather definite saving in internal
computing time as compared with the 7050 To use this additional speed
economically means mOre input and output units to be handled in parallel;
because IBM has been promoting the use of two mcs with the 705 as a means
of souping up its data transfer rate, a comparable technique must be employed with this equipment. Consequently, for business purposes it is
considered that a minimum of two Data Synchronizers will be required.
Even here, the economy of a 709 as compared with the slower but cheaper
705 must be kept in mind. It is not at all certain that the minimum con-
Original 2/15/57
5152.3
5152.3
(Continued)
figuration is economically justified against a 105 of approximately equal
production capabilities.
Type
Type
Type
Type
Type
2 Type
4 Type
16 Type
Type
Type
Type
109
136
141
146
737
166
755
129
111
121
117
-----
· · • ·• $ 10,000
·
·
1,100
· ······
·
1,400
·
·
·
·
·
·
1,300
·
····
1,200
· · · · · · 7,000
·
·
7,200
·· ·· ·· ·· ·
10,400
800
600
....···· ··
1,800
Total Monthly Rental.
· • $ 48,800
·• ·
Analytical Control Unit.
Power Supply Unit •
•
Power Supply Unit (Model 2)
Power Distribution Unit.
8,192-Word Magnetic Core Storage.
Data Synchronizers
•
Tape Control Units.
Magnetic Tape Units
Punch Card Reader (Model 2) •
Card Puncho
Printer With Type 757 Control Unit.
This minimum configuration does not provide a particularly large memory.
Punch card input and output is not efficient, requiring program conversion
as a part of internal computation, and probably would not be suitable if
any significant volume of either type operation existed. Printing is on
an off-line basis only, at a rated 150 lines per minute, or a maximum expectancy of about 50,000 net lines per 8-hour shift. Thus the installation is not especially attractive.
5152.4
Typical.709 Installation, Business Purposes. An equipment configuration
which appears more suitable for a large-volume business data handling
operation, of a scope for which the 709 capabilities may be attractive,
is as follows:
Type
Type
Type
Type
Type
3 Type
6 Type
20 Type
Type
Type
Type
Type
2 Type
709
736
741
746
738
766
755
129
160
730
714
722
727
·
·
·
· ·
-------
· · · ·· · · ·
·· ·
· · ·· ·· · · ·
·· ··
·· · · · . · · ·
·· ·· · .
··
$ 10,000
Analytical Control Unit.
Power Supply Unit.
1,100
Power Supply Unit (Model 3)
1,600
Power Distribution Unit.
1,300
32,768-Word Magnetic Core Storage
20,000
Data Synchronizers.
10,500
•
Tape Control Units.
10,800
13,000
Magnetic Tape Units
1,850
Record Storage Unit.
• •
2,775
High-Speed Printer (off-line only).
Punch Card Reader With Type 759 Control Unit.
2,400
Card Punch With Type 758 Control Unit.
1,050
1,100
Magnetic Tape Units (for off-line equipments)
Total Monthly Rental.
· · · • · · · · $ 17,475
This obviously is not cheap, but also can be shaved in several areas.
However, a range of $60,000 and up monthly appears to be a reasonable
conglomeration of units for this system. It is rather evidently priced
well above anything else so far announced in the business data processing
field and, for this reason, is not strictly competitive with either the
Univac II o.r IBM's own Type 105.
Original 2/15/57
5152.5
5152.5
IBM Announcements of Type 709 Costs. In its publicity releases and advertising on the 709, IBM has been citing monthly rentals of around $55,000
as "typical" costs of an installation. From the comparative rentals developed in the preceding two paragraphs, it is rather obvious that IBM's
"typical tt 709 is, in some respects, a quite stripped-down data processor.
For one thing, the input-output array of 5152.3 is just about as minimum
as can be visualized, costwise, for a commercial EDPM; another $6,000
worth of equipment, which would bring it up to $55,000, won't add up to
much. It c.ertainly won't admit of a 32,768-word memory, which is one of
the plugged features of the equipment. In some respects, the publication
of this figure has the appearance of a "loss leader;'~ get the customer's
order at all costs and then gradually break the news that another ten to
twenty thousand a month ought to be added to get a decent installation.
Original 2/15/57
5153
5153
OPERATING FEATURES OF TYPE 709 EDPM
This section is presented in preliminary form only and simply highlights
some of the features of the 709 which have been added to make it more
adaptable to business data processing. A complete revision of this material will be forwarded shortly. This section does not include the
characteristics of the Type 766 Data Synchronizer, which is written up
in detail in 1160. Characteristics of the Type 729 Tape Unit are included in Section 1105.
5153.1 Magnetic Tape -- Memory Transfer Rates. IBMos advertising and promotional
literature make it appear as if the tape speed or transfer rate has been
improved over that available in the 702, 704 and 705. Nothing of the kind
has developed, and nothing new has been added with the sole exception of
the verification of tapes immediately after writing. Their statements
of tape holding the equivalent of 350 decimal digits per inch, with a
transfer rate of 25,000 digits per second, are nothing but another way
of stating what can be done with the existing Type 704. The explanation
is simple. If information is in binary form internally and written on
tape in that fashion, then the 1200 information bits per inch of tape (6
bits per character, plus check bit, at 200 characters per inch) is the
equivalent of about 350 decimal digits, provided practically all bits
represent information. This will not always be the case; the internal
word is 35 bits plus sign, which is equivalent to about 11 decimal digits.
Unless typical numeric elements of business information (which average
much less than 11 digits) are packed two or three into the word, there
may be only a few--lO or 20--significant information bits. Consequently,
the statement that tape holds the "equivalent of 350 decimal digits" is
the maximum .cont'l'ition; in practice, it will practically always be less,
and often a whale of a lot less. The 729 tape units used with the 709
still pack at 200 to the inch, still move the tapes at 75 inches per second, and still handle exactly 15,000 tape characters per second maximum
transfer rate.
5153.2
Indirect Addressing. A major change from the 704 is the inclusion ofa
form of "indirect addressing" in an instruction; many, but by no means
all, instructions can be so treated. An example will serve to illustrate
how it operates.
Suppose memory address 1234 contains the number 1111.
single-address instruction of the form
ADD
In the normal
1234
the EDPM goes to address 1234 and takes its contents, 1111, to add to some
other quantity. In "indirect addressing," the contents of 1234 are not
added, but instead used as the location of another number which is the
quantity actually used. In this example, instead of adding "1111," the
709 would go to location or address 1111 and actually add the quantity in
that location. This is the meaning of "indirect addressing;tf, the address
attached to the instruction is used only to find the actual location for
the operand really involved. In the 709, the indirect addressing use of
an instruction adds 12 microseconds to the normal execution time.
Original 2/15/57
5153.2
5153.2
(Continued)
A major use of this feature is in the sorting and merging of raw data.
The conventional methods of sorting involve the constant reshuffling of
the items internally, by means of memory transfers, to place them in
proper sequence. With "indirect addressing," the items themselves are
never moved. Instead, a fttable" of addresses of each index number in a
group of unsorted items is set up in memory, and the comparison instructions refer to the addresses of each word of the table, not of the data.
The table itself is sorted into the sequence of the items, but this involves only one-word.memory transfers rather than the movement of long
strings of information. When coupled with the use of multiple control
words in the Type 766 Data Synchronizer, see 1160, the entire sorting
operation, including writing out of a long sequenced string of items, is
done without ever moving the data from the original input locations. The
"table" words are so set up in format that they are used directly as Control Words in the output operation. As an additional note, this technique
makes it feasible to sort items of different lengths, rather than treating
each item as a constant number of words or characters identical with all
other items.
Indirect addressing may be used with or without index registers applying
to the same instruction.
5153.3
Conversion Instructions. Thre~ new instructions have been added to the
equipment which permit the conversion of information from one number or
character code system to another. The specific and most obvious intent
1S to facilitate conversion from one system of character codes to another
or from decimal numbers to binary numbers and the converse. The method
of approach is sufficiently general to permit conversion from almost any
type of character code system to any other; for example, Univac to 705
codes or vice versa.
The three instructions all involve the use of tables. The general use
of the instructions follows the principle outlined: Each 709 word is
treated as six 6-bit code combinations. Each 6-bit group is added to
a base table address, in binary addition, to create a new address within
the table. The contents of this new address are two-fold: (1) a 6-bit
group which is the new code combination corresponding to the old one; and
(2) a new base address in the table from which to begin the addition for
the next 6-bit code combination in the word to be converted. The new base
address may be the·same as the, original one, or completely different, thus
permitting repeated references within a single table or a "bootstrapping"
operation through a complex, segmented table. This is the basic technique of conversion by replacement, the purpose of two of the three instructions, and results in a new word of six 6-bit code combination.
As an example of how this works, let's convert a "$" in Univac character
code to the same symbol in the 709 code~ Omitting check bits, the Univac
character code for "$" is 10 1101, which as a 6-bit binary number is the
saIIle as decimal "45". The 709 internal code for U$tt is 10 1011. Now let
us'set up in memory a table beginning at location 1000 and containing 64
words, the maximum number of code combinations in each of the two types
Preliminary Editi~n
Original 2/15/57
5153.3
5153.3
(Continued)
equipments. (Decimal notation will be used in this explanation for memory
addresses, although the actual operations in the machine occur, of course,
in binary). In location 1045 is a word which consists of two parts: (1)
its left-most six bits are tllO 1011," which is the 709 internal code for
"$;tt and (2) its rightmost bits are the number t'lOOO" (in binary, of
course).
By a separate instruction, the word in Univac character codes is placed
in the MQ register. Now a ttConvert by Replacement from MQu instruction,
with address 1000, acts as follows: The low-order six bits of the word
in the MQ register are added to 1000; with a dollar sign in this position,
the result is 1045.· Then the contents of location 1045 are obtained and
the left-most six bits placed in the left end of the accumulator. The
rightmost bits reset the starting point back to 1000. Now both the MQ
register and accumulator are shifted six bits to the right and the cycle
is repeated. This continues until six groups have been converted, at the
conclusion of which the accumulator contains, in proper order, the new
codes corresponding to those which were in the MQ. Provision is made for
terminating the conversion after any number of characters if fewer than
six are involved.
Two of the three conversion instructions operate in this manner and are
designed for character replacement. It is not necessary to return toa
pre-set address at the beginning of the table for each group of six bits
to be converted; each "word" in the table is used to designate the origin
point for the next character to be converted. It is at once obvious to
the cryptographically-inclined that this technique can be used to generate
virtually unbreakable codes, although the significance of this with respect to business information is not at all evident in other than the communications area.
The final conversion instruction operates by addition, and is the one
used to convert from decimal to binary and vice versa. In this case, 20
bits from the words in the table are added successively to the contents
of the accumulator to arrive at a total.
The conversion instructions are used with others in small sequences to
perform the entire cycle. IBM quotes these time figures for typical conversions:
Replacing leading zeroes by blanks in a six-digit coded word: Three
instructions and 144 microseconds total -execution time.
Same for a 12-digit (two-word) field: 168 or 336 microseconds, depending upon whether there are more than five zeroes to be replaced.
Converting six 6-bit character codes to another code system: Four
instructions and 168 microseconds total execution time.
Converting a six-digit coded number to binary form: Five instructions
and 204 microseconds total execution time.
Converting a 20-bit binary number to coded decimal: Five instructions
and 360 microseconds total execution time.
Preliminary Edition
Original 2/15/57
5153.4
5153.4
Index Transmission Instructionso The use of index registers in the 704
is subject to some restrictions which have been eliminated in the 709 by
incorporation of several new instructions to permit more flexible manipulation and use of the fib-box" technique. These instructions have been
incorporated along the lines of recommendations made by the SHARE organization, a group of IBM 704 users.
5153.5
Variable Length Multiply and Divide. The most time-consuming instructions
in almost all EDPMs are the multiply and divide operations. Rather than
always multiplying full words, or developing full word quotients, the 709
incorporates special instructions which permit the programmer to designate the number of bits involved in a multiplier or quotient and thus
eliminate the time required to pass through the unused bit positions. Although of use in business applications, which seldom involve multipliers
or quotients of the size of the II-digit (approximate) word of the 709,
the number of such operations is relatively few and overall time savings
are not too significant. However, it is a feature.
5153.6
Zero Skipping on Multiplication. Somewhat akin in objective to the previous feature is the modification of the multiplication operation to permit speeding up execution time by rapidly passing over zeroes in the
multiplier. If a multiplier is all zeroes, the answeris created in two
machine cycles, 24 microseconds. Normal multiplications vary from 72 to
240 microseconds, depending upon the number and distribution of zero bits
in the multiplier.
5153.7
External Signal and Floating Point Trapso In some types of on--line data
reduction work (real-time processing), it may be desirable to have information from an external source interrupt the program being processed
so that the new information can be handled immediately. An external signal meeting predetermined requirements will cause the 709 to complete the
instruction it is on and jump arbitrarily to memory address 0004 for the
next instruction. Simultaneously, the address of the instruction which
would otherwise have been next executed is placed in 0003.
The tffloating point trap" is somewhat akin to the automatic overflow
logic of Univac. If an overflow or underflow occurs in floating point
arithmetic operations, the address of the normally-next instruction is
placed in location 0000 and the 709 skips to location 0002 rather than
continuing in sequence. This trap is operative only in floating point
arithmetic and not in normal binary instructions.
5153.8
Sense Indicators. A new 36-bit register has been added to provide ~hat
amounts to 36 "selectors." Each position can be set to til" (on) or "0"
(off) and singly or in groups tested to provide alteration switches in
a pr09ram. They can be used also for extracting bits within a 709 word
or for permitting tests to be made on groups of bits within a word.
5153.9
Number of Instructions in the 709. The 709 has 189 instructions, conpared with 87 in the 704~ Of the 87, 84 are included in the 709 and perform identical functions, although the actual execution of some of them
is changed (i.e., in multiplying). The single "redundancy tape test" of
the 704 becomes six separate instructions, one for each of the six possible
Preliminary Edition
Original 2/15/57
5153.9
5153.9
(Continued)
Data Synchronizer channels. The "copy" instruction used for all inputoutput operations in the 704, is restricted to drums and cathode ray
tube in the 709. For all other devices, the Data Synchronizers and their
array of instructions have been added. Finally, the "transfer on MQ overflowtl of the 704 has been dropped; a new set of operational codes has
been added to handle this condition.
Of the 189 instructions, 42 are devoted to operations of the six Data
Synchronizer channels although they accomplish only seven different
functions. Each of the seven functions has six separate operation codes,
one for each channel. Similarly, a number of the other new instructions
added are essentially the converse of operations already possible; i.e.,
"Transfer on Storage Zero" and "Transfer on Storage Not Zero." These are
programming conveniences rather than separate operations. Most of them
have been added, however, as a result of suggestions made by the SHARE
organization.
Preliminary Edition
Original 2/15/57
5154.1
( 2)
5154.1
Instructions of the Type 709 EDPM
FIXED POINT ARITHMETIC
rID -._.-
1"0
! 0
0
..
Q)
'·03
f~~) ~ 15
-
.
..-{
~
~
Instruction
Description
~DD
24
B
X X ADD
(m) S, 1-35 + (AC)S,1-35~ (AC)S,-135
ADM
24
B
X X Add Magnitude
(m) 1-35 + (AC) 1-35 ~ AC l - 35
ACL
24
B
X X Add and Carry
Logical Word
(m) S, 1;..35 + (AC) p, 1-35 - ? AC p , 1-35
A carry from AC p adds into AC q and AC35 ;
a carry from AC q is lost.
SUB
24
B
XX
(AC)S,1-35 -
SBM
24
B
X X Subtract Magni tude
Subtr~ct
MPY 24-240 B X X Multiply
(AC) 1-35 -
(m)S,1-35-*ACS,1-35
(m) 1-35 ~ACl-35
(m)x (AC)~ACl_35 (most significant bits)
and MQl-35 (least significant).
Condi tions:
~R
~VH
If (m)= 0, O~AC and MQ
and proceed to next
instruction.
If (m);;z! 0, execution time
depends on significa6t
bits in MQ.
24-240 B X X Multiply and Roun ~ Executes a Multiply (MPY). followed by. a
Round (RND).
240 B X X Divide or Halt
If (m) > (AC): (AC) S,q,p,1-35 and MQS,1-35
':; .(m) S,1-35~MQ, remainder
to ~CS, 1-35.
If
DVP
240
B,"X
X Divide or Proceed
~ND
24
E - -
Round
(m)~
(AC), no division occurs, EDPM stops
with divide check indicator
on.
If (m) > (AC):
If (m)' (AC):
If (MQ) 1
=1:
If (MQ)l:=O:
Division occurs as above.
No division, but computer proceeds with Divide Check IndiGator on.
"111 is added to AC 35 •
No change in AC.
Original 6/1/57
5154.1
(3)
5154.1
Instructions of the Type 709 EDPM
(continued)
FIXED POINT ARITHMETIC
Q)
"0
o
(.)
VIM
~ ~
Time Q) Q)OH
(us) ~ ~ ~
24
E X -
Instruction
Variable Length
Multiply
(continued)
Description
(m),X(MQ)number of positions specified by count
Product is in AC S , 1-35 (most significant)
and in high order positions of MQ (least
significant). Remainder of MQ contains
original contents 'of high order positions
of MQ" ( AC s =- MQ S) •
Timing is dependent upon sequential
zeroes in MQ. If count is zero, instruction is interpreted as No Operation.
VIDH
24
E
~
-
Variable Length
Divide or Halt
If (m)
(AC):
(AC)S ,q ,p' ,1-35
and (MQ)S , 1~35
+(m)s 1-35~Qlow order positions
,
specified by count.
Remainder~ACS,1_35 and MQ (high order
positions minus count).
If (m) (AC): ~No division occurs, EDPM stops
with Divide Check Indicator on.
If count is zero, instruction interpreted as
No Operation
VDP
24
EX-
Variable Length
Divide or Proceed
Same as VDH, except that if ·(m) ~ (AC), nQ
division occurs, EDPM proceeds with
Divid~ Checl< Indicator on.
Origin~l 6/1/57
5154.1
(4)
5154.1
Instructions of the Type 709 EDPM (Continued)
FLOATING POINT ARITHMETIC
Time
( us)
~AD·
S4+
Q)
~
><
Q)
~
oM
'S '8
,:....·H H
Instruction
B X X Floating Add
Description
(m)"'" (AC) ---tAC and MQ.
1. MQ is cleared.
2. (m) -4.SR.
3 • IF (SR ) l-S (AC) l-S: . (SR ) and ( AC) ar e
interchanged. (Snaller characteristic
always is in AC before addition). If
AC p or q
1: Modification occurs and
<
==
4.
an incorrect answer results.
(AC) s ~(MQ) s.
5.
If the difference in characteristics> 63
O--+AC. Otherwise, (AC)9-35 shifts right
x positions (x is difference in characteristics).
Bi ts shi fted pas t AC 35 ~ MQ9
Bits shifted past M~5 are lost.
6.
(SR)l~S~(AC)l-S.
7-.
(SR)9-35+ (ACL~_35 ~ACS,9-35.
If signs are different, 1'5 complement
of AC is added.
Sa. If (AC) s
(SR) Sand /suml 1, a. carry
=
>
is added to ACS ; * (AC) 9-.35 .. and (MQ) 9-35
shifted right one place; 1-4 Al: 9 •
8b. If (AC) S~ (SR) Sand /sum/2 1 ,
(SR)S4AC S and MQS; carry ~rom AC9 is
lost.
If (MQ) 9-35 0: 1 added to AC35 •
=-
If ,(MQ) 9,35F 0: 2.' s complement of
(MQ) 9-35 ~ MQ9-35 •
J,um/
Sc. If (AC) SF (gl) sand
< 1:
, I' S. complement of (AC) 9-35 -4 AC 9 - 35 ~
9a. ]f (AC) 9-35= 0;
AC is cleared •.
9h. Results are normalized by shi fting left·
until AC q=- 1 •. Characteristic is 'reduced by one for each shift.
10.
(AC) 1-8 27·4MQl-S, unless (AC)= 0,
when zeroes~MQl_S.
Oriainal 6/1/57
5154.1
(5)
5154.1
Instructions of the Type 709 EDPM (Continued)
FLOATING POINT ARITHMETIC
Q)
"'CJ
8
Time
(us)
FAM 84
UFA
72
UAM
72
FSB
84
FSM 84
~
•
~
or-{
-~ ~ ~
J
STQ
24
B
X X Store MQ
(MQ)~m.'
SLQ
24
B
X X Store Left-Half MQ
(MQ)
, ,S' 1-17 -+mS ,1-17·
STZ
24
B
X X Store
+O~m.
STL
24
B X X 'Store Ins~truction
Loe atiJ9~,Couriter
(m) 18-35' unaltered.
,'"
Ze~;C)
Location of SIL instruction
+ 1~m21-3?·'
;:;C,': "':
Memory to Register:
CLA
24
B X X
Clear and Add
CLS
24
B X X
Clear
CAL
24
B XX
~DQ
2,4
B X X
Claax.and Add
'Lo,gtc~l Word
Load NQ
&
Subtrac,t
(,) ~ (AC S• l - 35 ;
'f":" ,(m) ---+ AC
,
," S,;,1,-35'•
(m)--+ ,ACp ,1"';35 ;
(m)
---7'
a ~ACq,p.
O~ACq,p.
O~ACS
'
,q •
MQ.
0'
Original 6/1/57
;.(
5154.1
(8)
515401
Instructions of the Type 709 EDPM (Continueg)
INDEX TRANSMISSION INSTRUCTIONS
Memory to Index:
0
0
Ti.me
(us)
LXA
24
0)
'1'j
e'
I><
Q)
0)
!:I
$ Hrg Hrg
B - -
LXD
24
B
--
LAC
24
B
--
LDC
24
B
--
Instruction
Load Index from
Address
Load Index from
Decrement
Description
(m) 21-35 ~ Index.
,
(m)3~17 --+ Index.
Load Complement of 2' s complement of (m) 21-35 ~ Index.
Address in Index
Load Complement of 2' s complement of (m) 3-17 ~ Index.
DecrVt in Index
Instruction to Index:
~xr
24
B
--
AXC
24
B
--
Address to Index
True
Address to Index
Complemented
21-35 of instruction in m~ Index.
2 v s complement of 21-35 of instruction
in m~Index •
Accumulator to Index:
PAX
24
B
--
PDX
24
B
- -
PAG
24
B
--
PDC
24
B
--
Place Address
in Index
Place Decrement
in Index
Place Complement
of Address in
Index
Place Complement
of Dec:rement
in Index
(AC)21_35~Index.
(AC)3-17 ~Index.
2 v s complement of (AC)21_35-+Index.
2's complement of (AC)3-17 --+Index.
Index to Memory:
--
Store Index in
Address
(Index).~
-
Store Index in
Decrement
(Index)~
SXA
24
B
SXD
24
B-
m21 - 35 ..
m3-17
1l
Index to Accmulator:
IPXA
24
B
--
!PXD
24
B
--
',Place Index in
Address
Place Index in
Decrement
O~AC;
( Index) 4
o -?AC;
( Index) ~ AC3 _J. 7
AC21-35 •
0
5154.1
( 9)
515401
Instructions of the Type 709 EDPM (Continued)
INDEX CONTROL INSTRUCTIONS
.
0)
ro
0
0
0)""'"
0)
~ .~
.~ ~ ~ rg
]
A- -
Instruction
Description
E-t- E-t H
rrIX 24
IINX 24 A
--
lIm 24
A
--
IrXL
A
--
24
If (Index» Decrement; (Index)- Decrement
-+ Index; control ~ mo
If (Index)iDecrement: (Index) unchanged,
control maintains normal sequence.
Transfer on No Index
If (Index)~ Decrement: (Index) unchanged,
control 4 m.
If (Index» Decrement: (Index)- Decrement ~
Index; control maintains sequence.
Transfer on Index High If (Index) Decrement: Control---} m.
If (Index)~ Decrement: Control maintains
sequence.
Transfer on Index
If (Index)~Decrement: Control~m.
Low or Equal
If (Index» Decrement) Control maintains
sequence
( Index )+ Decrement ---+ Index; Control ~ m.
Transfer with Index
Incremented
2's complement of location on TSX instrucTransfer & Set Index
tion ~ Index; control-+ m.
Use: Address modification by Index; address is reduced by contents of Index.
(Subtracting 2 v s complement is equivalent
to addition of oriainal value).
Transfer on Index
>
0
II XI
24 A
IISX 24 B
--
.-
CONVERSION INSTRUCTIONS
24
C
--
Convert by Replacement
from Accumulator
(AC~,1-35
handled as 6-bit character code
representations (right to left).
(AC)30-35 added to instruction address in
SRo Contents of this new address ~ SR.
(AC) shifted right 6 bits. (SR) S,1-5--+
AC p ,1-5. Count is decreased by 1. Steps
are repeated until count eauals zero.
CRQ 24 C
Convert by Replacement Similar to above, using MQ. Rotation is
from MQ
left to riqht.
CAQ 24 C - - Convert by Addition
(MQ1s, 1-35 handled as 6-bit representations.
Same procedure as CRQ above, except (SR)
from MQ
are added to AC q ,p,1-35·
Uses: CVR replaces a word in BCD form with a word in another BCD form (code
conversion)
CRQ is used for format control; ioe., to replace leading
zeroes with blankso CAQ is used for decimal to binary and binary to
decimal conversion.
PlR
--
0
Original 6/1/57
5154.1
( 10)
5154&1
Instructions of the Type 709 EDPM (Continued)
SENSE INDICATOR INSfRUCTIONS
Accumulator ~ Sense Indicator
PAl
24 B - -
OAl
24
B - -
RIA
24
B.;.-
IIA
24
B - -
PIA
24
B - -
(Instructi.on bits 12-35 not used)
Place Accumulator in
Indicators
OR Accumulator to
Indicators
(AC) p, 1-35
--+ SIO-35·
(AC)p,1~35matched with (SI)0-35.
Logical
"OR n operation -..... SI • Example:
(AC): 001100001100001100000000111111000000
(SI) : 000000111111000000001111111100000111
before
(SI): 001100111111001100001111111111000111
after
Reset Indicators from (AC)p,1-35 matched with (SI)0-35~
Accumulator
A ttlft bit in AC sets corresponding position
in SI to ftO·t.
A "Oft bit in AC leaves corresponding position
in SI unaltered. Examples:
(AC): 001100001100001100000000111111000000
(SI): 000000111111000000001111111100000111
before
(SI): 000000110011000000001111000000000111
after
Invert Indicators
(AC)p,1-35 matched with (SI)0-35.
A ttl tt bit in AC complements corresponding
from Accumulator
bit position in SI.
A t'O" bit in AC leaves corresponding position
in SI unaltered. Example:
(AC): 001100001100001100000000111111000000
(SI): 000000111111000000001111111100000111
before
(SI): 001100110011001100001111000011000111
after
Place Indicators
in Accumulator
Memory ;-+Sense Indicators
(m) 8.1-35 ~ 81 0- 35 •
OR Stor age to
~ame as OR Accumulator to Indicators, using
Indicators
(m) instead of (AC).
Reset Indicators from Same as· ttReset Indicators from Accumulator, tt
Storaqe
usinq (m) instead of (AC)
.
Invert Indicators
Same as 91 Invert Indicators from Accumulator, t.
f:rom storaqe
usinq (m) instead of (AC).
Storage Indicators
(SI )0-35 .--.., mS.1-35 •
LDI
24 B X X Load Indicators
OSI
24 B X X
RIS
24 B X X
lIS 24 B X X
SII
24 B X X
Original 6/1/57
5154.1
( 11)
5154.1
Instructions of the Type 709 EDPM (Continued)
SENSE INDICATOR INSTRUCTIONS (continued)
Instruction
(i)
rc:J
0
,0
(i)r-.
.~ ~
8-":"""
~ ~
><
(J)
(J)
~]
$.t
Instruction
.r-!
~
H
-D - D--
SIL
24
SIR
24
RIL
24
RIR
24 D -
IlL
24
D
--
IIR
24
D
--
Sense Indicator (Instruction bits 18-35 are control or F-field)
D
Set Indicators of
Left Half
Set Indicators of
Riaht Half
Reset Indicators of
Left Half
-
Reset Indicators of
Riqht Half
Invert Indicators of
Left Half
Description
F-field matched with (SI)0-17;
loaical OR~ SI.
F-field matched with (SI) 18-35;
looical OR ---+ SR.
F-field matched with (SI)0-17 a (1) bit in
F-field resets correspond1n~ bit position
of SI to f'O _tI
As above, on (SI)18-35-
F-field matched with (SI)0-17; a ttl tf bit in
F-field complements corresponding bit
jlosition in SI.
Invert Indicators of As above, on (SI)18-35Riaht Half
Transfer and Testing of Sense Indicators
TIO
24 B X X Transfer When Indicators On
TIF
24 B X X Transfer When Indicators Off
24 B XX On Test for Indicators
ONT
OFT
LNT
RNT
LFT
RFT
24 B X X Off Test for Indicators
4.8 D -.
Left Half Indicators
ON Test
48 D
Right Half Indicators
ON Test
...
48 D
Left Half Indicators
OFF test
48 D
Right Half Indicators
OFF Test
'
--
--
\
(AC) matched with (SI). A ttp' bit in AC
examines corresponding bit position in SI.
If all examined positions of SI are l's,
control ~ ffi; if not, next instruction in
seauence.
As previous instruction; if all examined bits
in 51 are OVs control~m.
(m) matched with (SI), in same manner as two
previous instructions .. If all examined
positions are l's, skip next instruction.
As previous instruction; if all examined bit
positions are O's, skip next instruction.
F-field matched } If all examined bit
with (SI)0-17
positions are l's,
skip next instruction
F-fie ld m~atched
with (SI)18-35
F-field matched
If all examined bit
with (SI)0-17 . > positions are O's
skip next instruction
F-field matched
with (SI)18-35
1
1.
Original 6/1/57
5154.1
(12)
Instructions of the Type 709 EDPM (Cohtinued)
5154.1
LOGICAL BIT MANIPULATION INSTRUCTIONS
(1)'-"
(I)
.~ ~
'"0
0
8 ........
0
CIM 24
COM 24
CHS
SSP
SSM
ANS
(I)
><
~
.1""\
~ ~
H
~A
E* E* E * E* iE * -
24
24
24
48 B X X
Instruction
Description
Clear Magnitude
0--+ AC a ,p,1-35
Complement Magnitude Complement AC q ,p,1-35
Change Sign
Set Sign Plus
Set Sign Minus
AND to Storage
ANA 36 B X X AND to Accumulator
Change sign of AC
Make sign of AC plus
Make sign of AC minus
(AC)p,1-35 matched with (m) S,1-35.
If corresponding bits of BOTH AC and memory
are 1 v s, 1~m.
If either bit is a ftO··, o---+om.
Same test as above; result ---+ AC p , 1-35
O~ACS,q-
ORS 24 B XX OR to Stor age
ORA 24 B X X OR to Accumulator
ERA 36 B X X Exclusive OR to
Accumulator
(AC)p,1-35 matched with (m)S,1.35·
When corresponding bit of EITHER AC or
memory is a ttl, ft 1---:,. m.
When corresponding bit of BOTHAC and
memory is a zero, O-..+m.
Same test as above; result --+- AC p , 1-35
0-+ ACs,q(AC)p,1-35 matched with (m)S,1-35.
When corresponding bits of AC and memory
are equal, O~AC.
When corresponding bits are not equal,
l~AC_
*-
Indexing not prohibited, but may cause change in the operation_
Original 6/1/57
5154.1
(13)
5154.1
Instructions of the Type 709 EDPM (Continued)
TRANSFER OF CONTROL INSTRUCTIONS
Unconditional Transfers
!> - 0).
Conditional Transfers -- Indicators
If overflow indicator ON, turn off and
control ---4 mco
If overflow indicator OFF, control~m;
Transfer on No
if ON. turned OFF, but seqUence maintained.
Overflow
Transfer on Overflow If Floating Point Indicator ON, turn OFF and
in Floatinq Point
control ~ m.
Transfer on Underflow If Floating Point Underflow Indicator ON,
in Floating Point
turn OFF and control ~ m.
Transfex on Overflow If Floating Point Overflow and/or Underflow
Indicator ON, turn both OFF and control--) m
or Underflow in
Floating Point
If Divide Check Indicator is ON, turn OFF·
Divide Check Test
and maintain instruction sequence.
If OFF. skip next instruction.
TOV
24 B X X Transfer on Overflow
TNO
24 B X X
TOF
24 B X X
TUF
24 B X X
TOU
24 B X X
OCT
24
E
*-
Original 6/1/57
5154.1
(14)
5154.1
Instructions of the Type 709 EDPM (Continued)
TRANSFER OF CONTROL INSTRUCTIONS (continued)
Conditional Transfers -- Comparison
Ql
rc
0
0
<
-.-I
(i)
~ ] H'E
-
Instruction
BTT
24
EX
ErT
24
E X - End of Tape Test
lOT
24
E-
TAO
·..
TFO
B X X Transfer on DOC A}
in Operation
24 B X X Transfer on DSC F
in Operation
TAN
24
-
Beginning of Tape
Test
Input~utput
Test
24
B X X Transfer on DOC A }
Not in Operation
• • co
TFN 24 B X X Transfer on DSC F
Not in Operation
X X Transfer on DSC A '
Redundancy Check>
X X Transfer on DSC F
Redundancy Check..
X X Transfer on DSC A"
End-of-File
>
X X Transfer on DSC F
End-of-File
J
X - Backspace Tape
TAR
·..
24
B
TFR
24
B
TAF
24
B
TFF
24
B
BST
24
B
BSF
24
BX
·..
REW 36
.. ~o
B X· ...
'~'iBackspace
~.Rewind
File
Description
Indicator for selected DSC
is tested.
If ON: Turn off, maintain instruction
sequence.
If OFF: Skip next instrlJction.
End-of-Tape Indicator for selected DSC is
tested.
If ON: Turn off, maintain instruction
sequence.
If OFF: Skip next instruction.
(Address of BTT and ETT instructions specify
particular DSC involved).
Beginning-of~tape
Input-Output Indicator (Address 00005) tested.
If ON: Turn off, maintain instruction
sequence ..
If OFF: Skip next instruction.
If selected DSC is in operation,
control~m.
If not in operation, maintain sequence.
Six instructions in this group.
If selected DSC is not in operation,
control ---+ m.
If in operation, maintain sequence.
Six instructions in this group.
Test Tape-Check Indicator for selected DSC.
If ON: Turn OFF, control --+ m.
If OFF: Maintain instruction sequence
Six instructions in this group ..
Test End-of-File Indicator for selected DSC ..
If ON: Turn OFF, control---+ m.
If OFF: Maintain instruction sequence.
Six instructions in this qrOUD.
Specified tape is backspaced one record.
,
Specified tape is backspaced .until end-offile indicator or load point is reached.
Specified tape is rewound to starting (load)
point.
Original 6/1/57
·:':1·':
5154.1
(17)
Instructions of the Type 709 EDPM (Continued)
515401
INPUT-OUTPUT INSTRUCTIONS (Continued)
Sense Instructions
,......,
Q)
"tl
8
Q)0l
.~
E-i
PSE 24
'NSE
24
Q)
~
CD
.~
1& 1l4 ~
E
E
--
--
Description
Instruction
Plus Sense
Minus Sense
Console
printer
cording
0140
O141}
to
0144
0161}
to
0166
1341}
1342
sense switches (lights) and punch or
hubs are tested and manipulated, acto address part of PSE instruction:
All Sense Lights turned OFF.
Specified Sense Light turned ON.
Corresponding Sense Switch tested.
If ON: Skip next instruction.
If OFF: Maintain sequence
Impulse sent to exit hub of punch
control panel of DSC A.
(3341-2 and 5341-2 same for DSC C & E
1360} Entry hub of printer control panel of
3360
DSC A,C,E is tested.
5360
If :i.mpulse present: Skip next
instruction
If no impulse: maintain sequence
1361} Impulse sent to exit hub of printer
to
control panel of DSC A.
1372 3361-72, 5361-72 same for DSC C, E.
Console Sense Lights are tested; address
(0141 to 0144) determines which one.
If ON: Turn OFF, skip next .instruction.
If OFF: Maintain instruction s.equence
Sense switches are equivalent to Alteration Switches on the 702/705 and similar to
BIeakpoint Switches on Univac.
Original 6/1/57
5412.2
Instructions of the DATAmatic 1000 EDPM.
5412.2
ARITHMETIC INSTRUCTIONS
All operands and results are handled as II-digit plus sign numeric words.
Code
ADD
SUB
MUL
Cycles
1
1
8
8
1
35
DIV
1
31130
SST
1
7
Instruction
Add
Subtract
Multiply
i
I
Description
(A)+ (B) ~ c.
If overflow, control
(A)-(B) .......
-.+
1988 (automatic).
c.
If overflow, control--+ 1988 (automatic).
(A))( (B)~ C (rounded high-order half of
22-digit product)
---+ t'Rem:ai.ndeI" register (1995)
(low-order half).
Divide
(B)+ (A) --4 'c (most significant 11 digits)
--+ t'Remainder" register (1995)
(remainder)
Substitute
Extract (A) and (C)/ into C.
If a bit of B is tt~,., insert corresponding
bit from A;
If a bit of B is "a," preserve the corxesponding bit in C.
Extraction is bit by bit, but may operate
on digits or characters by proper setup
of the ttextractor" word in B.
Original 6/1/51'
5412.2
(2)
5412.2
Instructions of the DATAmatic 1000 EDPM (Continued)"
SHIFT INSTRUCTIONS
In these instructions, A and Cdenote high-speed memory locations and B denotes
the amount of the shift. (A) .is always the word shifted; (A) remains unchanged.
Code
SLP
1 7+13
Shift Left, Preserving Sign
ffiP
1 7+ B
Shift Right, Preserving Sign
SLA
1
7t!;B
7t1 8
Shift Left,
Alphabetic
Shift Right,
Alphabetic
(A) shifted left number of digtt spaces
designated by B,.keeping the sign unchanged andunshifted. NUmeric zeroes
fill in at ri~ht. Result --.... C.
O~ B~ 12.
(A) shifted right number of digit positions
indicated by B, keeping sign unchanged am
unshifted. Numeric zeroes fill in at
left. Resul't -.. C. a ~ B 5 12.
(A) shifted left the number of
alphabeti~
spaces designated by B. Resul t ~ G.
o ~ B ~ 8. (If B=O, one two-bit shift
occurs).
.
As above, shifting right.
ffiA
1
SLN
1 7+B
Shift Left Numeric,
Sign Included
SiN
1 7+"B
Shift Right Numeric, As SiP, including sign position.
Sign Included
...
As SLP, including sign position.
O~
B ~. 12.
05
B~
Original 6/1/57
12.
5412 .. 2
(3)
5412.2 Instructions of the DATAmatic 1000 EDPM (Continued)
BUFFER-MEMORY TRANSFER INSfRUCTIONS
Instruction Word Format No. 20 Address A is the first memory location from or
into which words are transferredo Bl and B2 must be in the range 1-32. Address
C is used for the address of a subsequence call, if applicable.
Each basic order can apply to either the A Or B section of the buf-fer. The third
letter of the mnemonic operation code designates one of the four possible conditions:
A - Transfer from the A section of the Input Buffer~
B - Transfer from the B section of the Input Buffer.
S - Transfer from the ~ section of the buffer to which the last previous
connection was made (by a Transfer In, Double Transfer and Select,
Transfer and Select, or Read instruction)G
D - Transfer from the opposite section of the buffer to which the last previous
connection was made.
Transfer In Instructions:
Code
TIA
TIB
TIS
TID
Cycles
Instruction
Description
2 4+B 2 i'- Transfer B2 Words
If buffer is still being loaded from tape as
from Buffer to
result of the last previous Read instruc2 4+B2
tion, hold up via the interlock until the
Memory With InterRead is completed.
2 4+B 2
lock
If Or when interlock is released, transfer
B2 words from Input buffer to memory, be2 4 +B 2 !,
ginning at memory Address A.
If B2 is greater than number of words in
buffer, fill out with Sequence Change Instruction words (A and B address void,
1985 in C)
Examine words transferred for sentinels.
Store first sentinel in Sentinel Regist~r ~
If no sentinel present, store a
(1997)
Pass Instruction.
Ignore Bl .•
Same as previous instructions, ignoring
2 4+B2 Transfer B2 Words
status of interlock. (This permits tran~2 4 .... B2
From Buffer to
ferring Input Buffer contents to memory
Memory Without
2 4+B 2
after
a Read Instruction but before the
Interlock
2 4-+B2
tape transfer actually begins)
~
0
0
TBA
TBB
TBS
TBD
0
Upon completing these instructions, the next is taken from Address C (subsequence
:;call) •
.
An error (erroneous weight count) in any word transferred from the buffer results
in an automatic subsequence call to address 19850
Original 6/1/57
5412.2
(4)
5412.2
Instructions of the DATAmatic 1000 EDPM (Continued)
BUFFER-MEMORY TRANSFER INsrnUCTIONS (Continued)
Double Transfer and Selection Instructions:
Code
DrA
DIB
DTS
DTD
2
2
2
2
6+B2 Transfer to Output
Buffer and From
6..... B2
Input Buffer With
6+B 2 > Selection and
Interlock
6+B2
1; Transfer B2 words, starting at memory
address A, to Output Buffer Storage. Do
not examine overflow for sentinels.
2. Transfer B2 words from Input Buffer to
memory, beginning at address A.
a. Delay the transfer, via the interlock,
until the buffer has been loaded as a
result of the last previous Read Instruction.
b. Store first sentinel in Sentinel
Register; if none, store Pass Order.
c. If B is greater than number of words
in buffer, fill out with Sequence
Change Instruction words.
3. Extract one digit from the Bl word in the
Input Buffer and add to the units digit
of Address C (with carry perIl,li tted). The:1,;
digit extracted depends on the word in the:
Extractor Register (previously loaded).
4. Store the original instruction (DTA, DTB,
DTS or DTD) in Select Order Register (1994).
50 Make subsequence call to instruction in
Address C location modified by Step 3.
6+B 2 Transfer to Output
Buffer and From
6+B2
>
Input Buffer With
6+B2
6+B 2
Selection and
Without Interlock
Same as the previo1js instructions, except
the status of the interlock is ignored in
Step 2a. The transfer is made at once.
I'
DBA
DBB
DBS
DBD
2
2
2
2
Transfer and Select Instructions:
TSA
TSB
TSS
TSD
BSA
BSB
BSS
BSO
6+B 2 Transfer from Input Same as DTA, .... ,DTD instructions, omitting
Buffer With SelecStep 1. (Nothing is transferred from
2 6+B 2
>
memory
to the Output Buffer).
tionand
Interlock
2 6+B 2
2 6+B 2
2 .6+B2 Transfer from Input Same as DBA, •• o,DBD instructions, omitting
2 6+B 2
Buffer With SelecStep 1 .. (Nothing is transferred from
>
tion
and
Without
memory
to the Output Buffer).
2 6+B2
Interlock
2 6+B2
2
Original 6/1/57
5412.2
(5)
541202 Instructions of the DATAmatic 1000 EDPM (Continued)
BUFFER-MEMORY TRANSFER INSTRUCTIONS (Continued)
Txansfer Out
Code
TXO
Instructions~
Cycles
2
Instruction
4+ B2 Transfer B2 Words
From Memory to
Output Buffer
Description
to the Output Buffer B2 words beginning at Address A.
If Output Buffer overflows, examine overflow
words for sentinels.
a. If present, store first sentinel word
in Sentinel Register (1997).
bo If no sentinel in overflow, store a
Pass Order in Sentinel Register.
Address C may be used for a subsequence call
if desired.
l~ B2.! 32.
Bl is not used.
~ransfer
INTERNAL MEMORY TRANSFER INSTRUCTIONS
TXI
3
3+2C 2 Transfer Internally C2 words beginning at address A to new location beginning at Address B. I i C2 i 320
Range of address specified by A and C or
Band C should not include both 0999 and
1000 or any speCial registers.
TTX
1
5
Twin Transfer
TXS
1
5
Transfer and Subseauence Call
Contents of Select Order Register ----+ A
and (B) --+ C.
This order is used with Double Transfer and
Select and Transfer and Select in sorting
operations.
(A) --+B; sequence call,--.+ C.
Neither B or C should be 1990.
COMPARISON INSTRUCTIONS
Numeric and Alphabetic Comparison Instructions:
LCN
1
7
ICN
1
7
LCA
1
7
ICA
1
7
Numeric "Less Than" If (A) i (B), control---+ C. Otherwise,
Comparison
next instruction in sequence.
Numeric Inequality If (A) 1- (B), control ~ C. Otherwise,
Comparison
next instruction in sequence.
In both LCN and ICN, +0= -0.
Alphabetic "Less
If (A) i (B), control--+ C. Words are
Than tt Comparison
treated as alphabetic.
Alphabetic Inequality If (A) =I=- (B), control-4-C.
Compaxison
I
Original 6/1/57
5412.2
( 6)
5412.2 Instructions of the DATAmatic 1000 EDPM (Continued)
TAPE CONTROL INSTRUCTIONS
In the table blocks, Words 1 and 2 occupy the Key Channel; words 3-62, the
Satellite Channels. Either may be in a Read or Write status independent of the
condition of the other.
In tape instructions, the three addresses have these meanings:
Address A: Units and tens digits indicate the tape unit (00 to 99).
Address B: Can be used for changing Sequence Counter (transfer of control).
Address C; Can be used for making a subsequence call.
~
Code
Cycles
Instruction
RFA
1
7
Read Tape Forward - A
RfB
1
7
Read Tape Forward - B
RFD
1
7
Read Tape Forward Alternate
RBA
1
7
Read Tape Backward-A
RBB
1
7
Read Tape Backward-B
RBD
1
7
RFK
1
7
Read Tape Backward Alternate
Read Tape Forward ,Key Channel
RBK
1
7
WFA
1
7
1
7
Read Tape Backward Kev Channel
Write Tape Forward
"
WFP
*-
Write Tape Forward,
Except on· Key
Channel
Description
Read one tape block forward into Section A
of the Input Buffer.
Read one tape block forward into Section B
of the InoutBuffer.
Read one tape block forward iQto section of
Input Buffer 0Qi read into by last previous read order (i.e., change sections).
Connect the section not read into memorY.
Read 9ne tape block backward into Section
A of the Inout Buffer.
Read one tape block bac~ward into Section
B of the Input Buffer.
Same as RFD ,_ reading tape backward
Put K~y Channel of designated tape unit inRead sta~u$ and Satellite Channels in
Write status.
Read Key Channel of one block, forward, into
Section B of Input Buffer.
Same as previous instruction-, reading tape
backward.
Place Key and Satellite Channels of designnated tape unit in write status.
Fill Output Buffer to 62 words., if not full.
Fillers are sentinel Sequence Change instructions.
Write one block of 62 words on taoe.
Place Key Channel in read status and Satellite Channels in write status.
Fill Output Buffer to 62 words, if necessa.ry.
Write one block on Satellite Channels only.
Computer time necessary to execute instructions. Tape time depends on state
of tape unit in relation to action required by current instruction.
Original 6/1/57
5412.2
(7)
5412.2
Instructions of the DATAmatic 1000 EDPM (Continued)
TAPE CONTROL INSTRUCTIONS (Continued)
Search Instructions:
General commentsl Input Buffer A is organized to include ten 2-word sections
addressable by the digits a through 90 The t·searcht. instructions following
permit reading the Key Channel only of a tape block into the section having the
same number as the units digit of the tape addresso Up to ten separate tape
uni ts may be 't/searchedt' simultaneously (a separate instruction is required for
each), provided the units digit of the tape address differs for each one. (Thus,
tape units 12 and 42 cannot be searched simultaneously, but 11 and 12 can)~
Code Type Cyco
SFR
1
7
SFW
1
7
SBR
1
7
SBW
1
7
REW
1
7
Instruction
Search Forward,
Reading
Search Forward,
Writing
Search Backward,
Reading
Search Backward,
Vi1::,iting
Rewind Tape Unit
Description
Place Key and Satellite Channels in read
status.
Read Key Channel only of one block, forward,
into matching section of Input Buffer A.
Place Key Channel in read status and
Satellite Channels in write status~
Read Key Channel only of one block, forward,
into matching section of Input Buffer A.
Same as SFR above, except read backwards
one block.
Same as sFw above, except read backwards
one blocko
Rewind the specified tape unit~ Upon completion, tape is positioned for reading
or writina on the first blocko
KEY COMPARISON INSTRUCTIONS
FKC
1
First Key Comparison
SKC
1
Second Key Comparison
1. Move the first word in the Input Buffer A
section designated by the units digit of
Address A to Address A. (Ioeo, if A is
"0936,'11 first word of section 6 is moved
to 0936)0
2. Compare (A) and (B). If (A)1. (B),
control~ C"
Comparison is alphabetic,
not numerico
3" If a previous ttsearch·' .is not completed,
action is delayed by interlock until Key
Channels have been read in. Only the
section involved is interlocked; others
may be used if not interlocked.
4 .. Interlocking occurs from a "search t' instruction; if not interlocked, (whatever
is in section designated is used.
Identical with above, except the second word
in the designated section of Input Buffer
A is transferred to Address A. (SteJ> 1)
Original 6/1/57
5412.2
(8)
5412.2 Instructions of the DATAmatic 1000 EDPM (Continued)
PRINT INSTRUCTIONS
Code
Type
Cyc.
Instruction
PRA
1
6
Print Alphabetic
PRN
1
6
Print Numeric
Description
(A) printed as 8 alphabetic characters on
console typewriter.
B may be used to change Sequence Register.
C may be used as subsequence address.
(A) printed as 12 digits on console typewriter.
Band C maY be used as in PRA.
CONTROL INSTRUCTIONS
PSS
1
6
Pass
OST
1
5
Optional Stop
"
,
SCS
1
6
Sequence Change
BAR
1
7
Branch and Return
No operation. Next instruction taken from
Sequence Register •. Addresses A, B, C
immaterial.
Stop, or not, depending on console switch
setting and (B). (B) is a 12-digit word
of numeric O's and l's. The optional
stop depends on the particular (B) and tm
setting of the switches. (B) are numbered
1 to 12, from left to right.
I f the digit in position 12 is a ·'1," STOP,
unconditionally, regardless of switch settings.
If positions 12 and 11 are both "0," setting
of ()ptional Stop Switch is next controlling factor:
If, set to "Stop,·' STOP.
I f set to "Proceed," PROCEED.
I f set to "Compare, t. STOP if any of the
digits 1 through 10 is a "1ft s.ru! corresponding Breakpoint switch is ON; if
none is on, PROCEED.
Change Sequence Register to Address B,
Subsequence Call to Address C. A is disreaarded but must be numeric.
Subsequence Call to Address C.
Change Sequence Register to Address B.
Store in Address A an order which can be used
later to change Sequence Register back to
its current reading. (This storage occurs
after Sequence Register has been incremented by "I.")
Original 6/1/57
8110
8110 BASIC PRINCIPLES OF FILE MAINTENANCE ON A MAGNETIC TAPE PROCESSOR
Univac I--the first of the modern electronic data processing machines
adapted to the processing of business data--and all of its large-scale
successors and competitors have magnetic tapes as the major medium for
the storage of file information. Some of the medium-scale machines
use only punch cards for this purpose--such as the IBM Type 650 in its
first two years of availability--but the current trend is for all generalpurpose business data processors to provide, at some stage of development,
magnetic tape storage; this is certainly true in the above-$lOO,OOO purchase price range.
Magnetic tapes con.sti tute the only form of mass information storage now
available. Access to this type of storage is sequential; that is, every
item in the file must be passed over to reach any specific one desired.
The other major type of storage device now being developed for EDPM use
is the random access storage, of which the first announced versions are
the IBM RAMAC disc files. This type of storage is differentiated from
the sequential magnetic tape files in that access to any given item is
essentially of the same time; it is not necessary to pass over unwanted
information in the manner required with tapese
The discussion which follows is concerned solely with the characteristics
of ha'ndling files which are on magnetic tapes.
8111 DEFINITION OF TERMS USED
There are no uniform standards in the computer industry~ or among users,
for the terminology applied to the various factors comprising a magnetic
tape file of information. Those used here are not proposed as such
standards, but in this section the various elements will be referred to
specifically, as defined below; in other sections some of the alternate
terms mentioned are used. The basic reason is that lack of standardization makes it desirable to know not only our own terminology, where it
is definitive, but also the terminology which is used by competitors for
the same factor.
A file is one or more magnetic tapes containing information pertaining to some business data processing function. As such, it is very
variable, depending upon what information is included. For example, one
EDPM user may have a file containing information necessary to process
payrolls and another containing personnel information, much of which may
be repeated in the payroll file. Another user might combine the two into
one file. Any magnetic tape input to or output from an EDPM is a file,
because they all contain information pertaining to some facet of business
data. A master f.!J&. is generally restricted to a major file of information which is used over and over again, the pieces of information being
being changed as necessary to reflect current conditions.
81ll.1~!:i.J&..
8111.2
~
An item is all the information pertaining to a given major entry
in. the file. The file itself is sequenced according to some identifying
information which distinguishes each item from all others in the file.
For example, the item may be all the constant (or reasonably so) information necessary to calculate an employeells weekly pay. Name, clock number
8111.2
8111.2
(Continued)
hourly pay rate, department number, year-to-date gross earnings, number
of departments or income tax table classification, various types of deductions, etc. Normally, this type of file is sequenced on clock number;
the file itself consists of one such item for every employee of the company. In IBM terminology, an Item is usually referred to as a record;
their term is also used with other meanings.
8111.3 Data Element. A data element is one of the pieces of information which
consti tute an i tem--clq.ck number, employee name, etc. It can be considered as a complete and independent unit of information.
8111.4 Magnetic Tape Record. A magnetic tape record is the number of characters
or words of information written between successive gaps (no information)
on tape. It may be a fixed number of characters or words, always constant
for a given equipment, or the number may vary, not only between files but
for successive tape records within one file. In Univac, a tape record is
always 60 words or 720 characters, and is called a~; in 705 usage,
several records are normally combined into one grouped record. Other terms
are also encountered.
8111.5 Fixed and Variable Word Lengths. In some EDPMs--such as Univac--all operations are performed on a specific number of characters, called a word"
Several short data elements may be combined into one word, with extraction
being required to separate out any specific one needed. In other EDPMssuch as the 705--operations are performed serially on one character after
another, separate provisions (other than a constant number of characters)
terminating the individual operations; i.e., an instruction may work on one
character or 100. For practical purposes, a word and a data element can
be considered as identical in a variable word length machine; they are not
in a fixed word length equipment, in which a data element may be only part
of a word or consist of several words.
8112
8112
CHARACTERISTICS AND IDENTIFICATION OF BUSINESS ITEMS
The basic unit of information upon which an EDPM-- or any other type of
data processor--operates is an item. It is difficult to imagine any sort
of productive work on a file which does not require searching it, in some
manner, for one or more specific items, and performing some sort of operations on one or more of the data elements which are, or can be, a part
of the item. For this reason, it is necessary to understand clearly the
general nature of business items and the various means which may be used
to distinguish them one from another. This section is devoted to this
subject.
8112.1 General Variability; of Business Items. It is to be expected that the
number of words (or characters) in items from different files would not
in general be the same; each must be long enough to contain the information necessary for its specific purpose. It is equally true that,
within a single file, items are of different lengths, and frequently the
amount of the variation is extremely largeo In other words, the typical
business file does not consist of items of constant length. To be sure,
such files do exist; in some, the nature of the information to be represented is such that every item contains an identical number of words. In
others--and this is probably closer to the truth--they are set up in constant length to meet the requirements of some data processing system.
This is an artificial consideration which is not inherent in the nature
of the information to be represented.
Example of files which contain items inherently variable in length are
easy to enumerate; those which contain items of constant, or nearly so,
length, much more difficult. The .most obvious case of the latter which
come to mind is a name and address file used for magazine subscription
fulfillment and mailing operations; here almost every item has the same
number of data elements--the basic variation occurs in coded information
representing conditions which may be useful for promotional purposes.
But consider the possibilities for variation in item length. A master
insurance policy file contains items of lengths varying with the type
of policy, with outstanding loans, with delinquencies in payment, etc.
A central inventory file has items whose lengths may vary widely depending upon the number of warehouses carrying the item, the number of back
orders attached to the item, the number of purchases due in, etce A
department store charge account file may have provisions for a detailed
listing of payments made, of charges incurred, and of other persons
authorized to make charges against a basic accounto In other words, the
mythical tttypical" business file is characterized by a marked degree of
variation in the length (i.eo, amount of information) of items within
the file ..
In general, a business file consists of a number of items each of which
consists of a fixed master portion comprising data elements common to
~or practically all, items, and to which are added or appended a
variable number of other data elements which may range in number from
none to several hundred.
811201
(1)
8112.1 (Continued)
There is nothing novel or revolutionary in this statement. We are all
familiar with Kardex files, in which the fixed master portion comprises
the visible information, plus possibly additional data elements, and in
which the number of line entries representing various types of transactions which have been posted varies from nothing to the capacity of
several cards. In punch card operations, the concept of "master" and
"detail1t cards is also commonplace; there may be several kinds of detail
cards in one file, the number often being determined by selector capacity
to distinguish them from each other and to provide for different kinds of
treatment. Also commonplace is the break-up of closely related information into several types of files, usually because of the limited processing capabilities of the machines; thus a payroll master card may be a
completely different punch card from a year-to-date earnings cards, which
carries cumulative totals or earnings and deductions. It should be noted
that the separation into two files is not inherent in the data, but is
made because of the restrictions of processing equipment
The typical business item--if there be such an animal-- then may be considered as a fixed master portion followed by a variable number of detail
sub-items, thus:
Fixed-lenqth Master Portion of Item
First Sub-Item #1
I
t
·
·
· #1
Last Sub-Item
I
First Sub-Item #2
··
Last Sub-Item #2·
·
·
·
I
I
.
I'
First Sub-Item #3
Last Sub-Item #3
and so on.
I
The various types of sub-items each consist, normally, of a specific
number of data elements comprising a specific number of words or characters. In general, each type is ofa length different from the others,
but at any rate its length is set by the amount of information which it
contains. Any given item consists of the master portion (always) plus
anything from none to some maximum number of each of the possible subitemsl the sub-items can occur, of course, in all conceivable combinations.
This sort of arrangement is not the most complicated which can exist.
It is perfectly possible for one or more of the sub-items to be of
variable length--in other words, to consist of some sub-master portion
plus a variable number of sub-sub-items. The principal is identical to
the one-stage process depicted above; the complexity of processing may
8112,,1
8112.1
(Continued)
( ?)
naturally be increased as "details are piled on details are piled on
details. ft_ But from the standpoint of what is necessary to represent
the information which must be contained within the item, there is no
reason we can't have sub-sub-sub-items within sub-sub-items within subitems within items. Within files, of course.
8112.2
Arrangement or Sequencing of Items Within Files. A file may consist
of from a very few to a large number of items--for example, the basic
wage file maintained by the Social Security Board exceeds 120,000,000
separate items or accounts. As soon as the number exceeds a handful, it
becomes convenient, if not essential, to arrangement the items comprising
the file in some logical sequence. The particular sequence into which
any given file is ordered is that which is most appropriate for its use;
it is quite common for two different files to contain essentially the
same information (data elements), but to be maintained in two different
sequences for some processing consideration.
The criterion or factor which determines the position of any given item
within the file is a data element in the fixed master portion, usually
referred to as an index or control number. Frequently it is created
and placed into the item as a data element used largely for sequencing
purposes; such index numbers are employee clock numbers, social security
acco~nt numbers, insurance policy numbers, etc., are of this nature.
Many of them are strictly numeric; one major reason for their prevalence
is that sorting (i.e., sequencing) of numeric index numbers on punch
card equipment is faster than on alphabetic numbers. Many such numbers
are of course, alphabetic or alphanumeric--names, part numbers, stock
numbers, etc.
The items within the file are placed in order according to the index
number associated with each. For this purpose, the digits, letters and
special characters which comprise the index are ordered in a specific
sequence, which frequently is related to a sequence recognized by some
of the processing equipment. The sequence of alphanumberic characters
in a Remington-Rand collator, for example, is different from that in an
IBM collator, and consequently a file with identical index numbers will
be in a different order if Remington-Rand equipment is used from what it
will be with IBM hardware. The specific sequence is somewhat immaterial,
so long as the processing equipment handles it correctly and always works
with the same ordering; communication between two different types of
processing installations may be somewhat complicatedo The number of
characters in an index number may vary greatly; they range from a normal
minimum of four or five up to 22 or 24 for a name-sequence file and occasionally exceed this number (i.,e., the standard descriptive number
used by the bearing industry is about 40 digits).
In some files, the nature of the establishment of index numbers precludes
the possibility of duplication; this is usually true or clock numbers,
insurance policy numbers, and other types of identifying numbers whose
assignment is controlled by the user. This is not always the case, and
duplication may occur; i.e., in a file sequenced on names. In. a file
where index numbers of two different items may be identical, there is
almost always a secondary index number which is used to arrange the items
in a specific order within the prime index. This may be carried through
8112.2 (Continued)
several successive stages to provide a specific ordering where primary,
secondary, tertiary, etc., index numbers may be identical. Thus, in a
geo-alpha file, the principal sequencing may be on state, the secondary
on city within state, the third on postal zone within city, then on name
within postal zone, then on street address and finally on apartment number--to account for two John Smiths living in the same apartment building
in New York 56, N. Y.
8112.21 Two Basic Principles of File Sequencing. Because they are fundamental in
the handling of magnetic tape files, it may be well to state two basic
principles of file sequencing.
(1) Every Index number which is, or can be, used to determine the proper
ordering of the items in a file must be contained in the master
portion of the item and each is assigned a specific and unchanging
level or stage in relation to the other index numbers.
(2) If two different items have identical index numbers through all
levels or stages which have been established, then the order of
the two items in the file is immaterial. Because this condition
is, in general, to be avoided, this statement may be considered
tantamount to saying sufficient levels of index numbers must be
established to assure that every item in the file differs from any
other in at least on index number.
8112.22 Sequencing of Sub-Items. Because an item may consist of a variable number of sub-items associated with a master portion, it is necessary to
provide some method of sequencing them in some orderly manner following (usually) the master portion. The same principles and concepts
used to order the items themselves are applicable here; the sub-items,
and their sub-sub-items, etc., are ordered in some logical manner on
index numbers (data elements) which are a part of the information in the
sub-items. Thus the entire file is sequenced or ordered in some more or
less logical fashion, according to clear-cut, exhaustive and unchanging
rules.
8112.3
Identification of Items. Because items frequently are of variable length,
it is necessary not only to know the order they are in, but also to have
some means of ascertaining when one item has been passed and the next
one reached. In a manually-maintained file, such as a Kardex file, this
is relatively simple~ each item is a separate pocket, even though there
may be several cards. In a punch card file, an item mayor may not be
contained in a single cardJ usually it isn 9 t. There are numerous conventional techniques of identifying the various types of cards by control
punches which differ for each sub-item and for the master portion. Further, for master portion or sub-items which may exceed the capacity of a
single card, it is common practice to number the cards in sequence, from
"In up, in addition to the control punch, to assure their proper ordering.
Such control punches and numeric codes become a part of the sequencing
cri teria," with a 'specific relationship to other index numbers; they are
among the data elements contained in the master portion or sub-items of
an item, although strictly speaking they are not inherent in the information necessary to be carried with the item. They are data elements
8112.3
8112.3
(Continued)
artificially introduced because of characteristics of the processing
system, not because of the inherent information requirements of the file.
Similar examples can be found with other types of data processors, such
as the notches in an addressograph plate.
The foregoing may be summarized briefly: . All business files are ordered
or sequenced in some logical manner, based upon a definite number of index
or control data elements which are a part of the items, the index numbers
having a specific order of importance. In addition, the master portion
and the various sub-items which comprise the item are identified, one from
the other, in some unique manner. The number of indices upon which a file
can be sequenced, the nature of the control symbols identifying the various
sub-items which comprise an item, and the amount of information which can be
associated with each often is predicated upon the characteristics of the
processing system used.
EDPMs handle business information in the same sort of files as the more
conventional types of data processors. Like them, the files must be
sequenced in some orderly and predictable manner, the various pieces of
an item must be identified, and the equipment must be able to find any
given item, or portion thereof. The basic principles of magnetic tape
files are identical with those discussed here; the techniques very
because the characteristics of the data processing equipment differ
from those used in the past. In the next section, consideration will
be given to some of the methods which can be used to establish magnetic
tape files; following that, the implications of the normal means advocated
for working with the file will be investigated.
8113
8113
EDPM REPRESENTATION AND IDENTIFICATION OF BUSINESS ITEMS
The basic logic for establishing the format for a file of variable-length
items in an EDPM magnetic tape system is identical with that for manual,
punch card, or any other sort of file, Similarly, the logical operations
involved in handling a manual file are identical with those for handling
a tape file. The techniques vary because the nature of the processor is
totally different, but exactly the same operations and conditions must
be accounted for in both systems.
8113.1
Characteristics of Magnetic Tapes and Processors Affecting File Formats.
Like any other type of data processors, EDPMs and magnetic tapes have
characteristics affecting the format of files to be processed. One of
the principal advantages of an EDPM, as compared with more conventional
processors, is its extreme speed and capacity for storing information;
rather than being limited to a few counters (as adding machine) or a few
score (as a punch card tabulator or calculator), it has the ability to
retain a large number of characters--20,OOO in the 705, 24,000 in Univac
II, and options in both equipments for still greater storage. They have
great flexibility in their aptitudes for transferring information around
inside the machine; among other things, this means that an index number
doesn't always have to be nin card columns 10-24."
The magnetic tapes have special characteristics, also. Just as Kardex
or punch cards, they are nothing more than a medium for storing information. At some point, the information contained in them is transferred
into the tlmemorytt of an EDPM, where it is an exact character-by-character
image of the tape record; there is an exact counterpart in the transfer
of information contained as holes in a punch card into the counters, relays, selectors, etc., of some type of punch card equipment. The EDPM
differs, in one important respect, in its ability to juggle this "tape
image" around as it sees fit.
Now consider the tape itself, and imagine a file of information stored
on it. Physically, it is a counterpart of a Kardex cabinet or a box of
punch cards. It has one important difference: It's a physical entity
which can't ttlos e'· a piece of itself. In this respect, it is. unlike the
'other two, in which it is possible to lose a card. It has other direct
comparisons. Just as a Kardex or punch card file can overflow a box as
new items are added, the tape can get .ttoo full;" the operation of taking a handful of cards and inserting them at the beginning of the next
tfbox" has its magnetic corollary in taking the end of one tape's information and writing it on the beginning of another"
But a tape isngt packed solidly with information from beginning to end.
It stores its information in records or blocks which are separated by a
blank space; the reason for this is that present-day memories, at least,
aren't big enough to hold everything we could pack on a reel of tape, so
the information stored is broken up into pieces of a size we can handle.
These pieces are called blocks or records, and their size varies with
different EDPMs. Compared with the infor-mation capacity ofa punch card,
a block is much bigger; compared with the capacity of a Kardex card, it
is usually smaller. In a punch card file, information is added and deleted by inserting and removing cards; sometimes by replacing with a new
8113.1
8113.1 (Continued)
card. The same thing occurs in a tape file; the big difference is that
the size of the block is bigger, and the additions and deletions often
mean "expandingtf or "compressing" an existing block rather than creating one of completely new information. But the principle is the same.
And, just as adding enough punch cards eventually necessitates reassembling the boxes holding the file, so adding enough information to a tape
means reassembling them.
From the standpoint of file formats, of course, one of the salient differences distinguishing them from punch card operations is the fact that
pieces of tape can't get lost; the processor doesn't need redundant information to identify successive sub-items and associate them with the
master portion; and, finally, the basic processing capacity and information holding ability of the EDPM are great enough to remove the limits
(at least for practical purposes) of how much information any single item
can contain.
8113.2
Tape Form~_ats. for Constant Length Items. If the items of a file are of
constant, or practically constant, length, they can be established in a
format of a specific number of characters or words; all items are then
of the same length. Each data element .is in a specific location or
"field" within the item. This is a direct counterpart of the typical
punch card layout, with the exception that the number of characters can
be much greater than 80 or 90. The size of the item does not need to be
an even sub-multiple of the tape block or record; a 17-word item in a
Univac causes no difficulty. The fact that some items will be split between two tape records does not mean that an identifying index number
must be repeated; the EDPM isn't going to·'lose u , a tape record and is
perfectly capable of holding two or three tape blocks in its memory.
This does not mean that the size of the tape block can be ignored completely in determining the most appropriate format; other functions of
which the EDPM may be capable should be considered. For example, the
multiple-tape "search'" feature of the DATAmatic system may make it advisable to place one item in each tape block, even though the entire 62
words are not required. Similarly, limitations of the high-speed memory
transfer in the IBM 705 may make it advisable to group the largest number
of full items which can be contained in 1,022 characters, and not attempt
to write the maximum size tape block possible. These considerations, of
course, are caused by characteristics of the EDPMs and not of the tapes
themselves.
8113.3
Tape Formats for Variable Length Items. More typically, the establishment of an item format for a magnetic tape prQcessor must recognize the
existence of a variable length item. Because the EDPM has no advance
knowledge of the length of any individual item, or of the types of subitems or the number of them that may be included, it is necessary to provide, in some manner for three factors: (1) Where does the item begin
in memory? (2) How long (i.e., how many characters or words) is it?
(3) What sub-item do'es it include? The answers to these three questions
must be contained within the item ftself, or must be capable of determination from something in the item. Thus, because the tape format is nothing but a representation of the initial transfer of the tape block into
8113.3
the central processor, it follows that the tape format must contain
the necessary information to answer these questions. The following
paragraphs describe some of the techniques which have been developed
or proposed.
First, however, it may be well to state specifically a factor which has
been implied in the previous sections: The development of an item format
for a magnetic tape file is identical with that of the layout of input
and output areas in memory.
8113.31 Word or Character Counts Contained in the Item. One method of keeping
track of variable length items is to include as a part of the master
portion a data element which is in the number of words or characters in
the item. This is placed in a fixed and constant position within the
master portion. As an example, consider an item format development for
a Univac tape file, It has a fixed master portion of eight words, and
may have up to 90 words or variable information appended to it; thus it
never exceeds 100 words in length. Two digits then can express the length
of any given item& For convenience, assume that they are the last two
digits in the third word of the master portion; which word is of no consequence as long as it is constant for all items.
81 variabl~~!...SU_b_-_I_t_e_m_ _ _ _....:
Fixed Master Portion of Item
Now suppose that this item begins in word 13 of a Univac block this puts
the word count in word 150 Since we designed the item layout, ,a knowledge of where the first word is located immediately enables us to find
the word which has the counter; the converse is also true. The word count
is a pair of digits indicating the number of words in the item; in this
case, anything from OS"(no sub-items attached to the master portion) to
98 (the maximum size p~5sible)o Suppose it is 23. This signifies that,
the next item begins 2~_words further along than the current item; because
the current item start'S at word 13, the next starts at 36 (25 f 13). Likewise, it's also indica~d b'y,! a word count, which is known to be in the third
word of the item, or in word 38
Thus the word count included as a part of
the master portion of the record can be used to modify addresses in an instruction sequence to enable us to i'bootstrap·· ourselves along the block
in memory.. Each item contains an indication of how long it is, and consequently tells us where the next item begins. It is not even necessary
that an item be contained completely within one block in the case of Univac,
which has a 60-word block, the calculation of the location of the next item
is always tested against 60; if the new address is greater than 60, the next
item is known to be in the next block. This is, of course, a programming
detail which really has nothing to do with the item layout, but is necessary
because the EDPM cannot read an entire file into memory at one clip.
Q
There are numerous variations of this basic technique., Because the master
portion of an item is always fixed in length, it is really not necessary
to include its number of words in the total ~ount; only the number of
8113.31
8113.31 (Continued)
words in the variable sub-items is required. The various types of
sub-items may all be of different word lengths, in which case a word
count in each sub-item present may be used not only for the bootstrap
operation, but also to identify the particular type represented. In
a file consisting of extremely long items, it may not be possible to
include the total word count in the master portion, because memory may
not be big enough to hold an entire item at once; obviously, the length
cannot be known until the entire item has been handled. In this case,
the word count may be calculated only to advance the bootstrapping to
the next block, which contains another counter to permit another advancement, and so on.
This method requires continual address modification to advance through
successive items in a file; the amount of computing time required to do
the modifications and test for the end of a tape record must be considered
in determining whether a single counter per item, or several, can be used.
Use of this method also requires that the word count be recalculated and
corrected every time the total length of the item is adjusted by addition or deletion of sub-items. It is, however, a commonly-used and
perfectly acceptable method of enabling an EDPM to keep track of where an
item begins and how long it is.
8113.32 Coded Item and Sub-Item Identifications. Another possibility is to carry
over into EDPM practice the basic logic of "control punches'· used in
punch card operations. In this method, the master postion and various
sub-items carry identifying characters in specific word or character
locations; these serve not only to identify the particular piece being
examined, but also enable a program to substitute the length of the particular piece into an address modification routine to permit "bootstrapping"
through the file. This is always possible, because the format allocates
a specific number of words to each type of sub-item or to the .master portion. Thus a coded key might be placed in the first character position of
the second word of each piece of the item:
I
~
~
~I
I
~
I
~
-
xl
I
zl
I
~
I
I
Here the tfAn identifies the master portion, 8 words long; the "X" the
first variable sub-item of 6 words; lty" the second sub-item of 5 words;
and"Z"the third sub-item of 4 wordso The method of using these codes
to progress though a file is somewhat different from the previous type
in technique, although the basic principles is quite similar. Suppose
the first "A lf has been located. This is known to be an 8-word master
portion and hence the next key code lies eight words further along.
-It is not known what it is, but tests of the code against constant A,
X, Y, and Z characters stored as a permanent part of an EDPM program
enable its identification, which in turn not only tells us what the item
is, but--since the various sub-item lengths were initialJy established at
a fixed number of words for each type--how long it is; i.e., where the
8113.32
8113.32 (Continued)
next piece of information begins. In the normal EDPM program, the
t'bootstrapping n requires more computing time, because the advancement
is by master portions and sub-items, rather than complete items. The
logic of the technique is quite straightforward and this is a standard
wrinkle in the repertoire of any good computer programmer. It is extensively used in Univac I installations, for example, although the
length of sub-items and master portions is usually either 2, 4 or a
multiple of ten words. The reason for this restriction has nothing to
do with the generality of the technique~ but is based upon the fact that
Univac I has only two and ten-word internal transfer orders.
Key codes can, of course, be combined with word or character counts of
identify the pieces of information which comprise a business file.
8113.33 Key Words. An approach rather different is the use of key words to signify which parts of an item are present. Unlike the previous two methods,
this is generally limited in its applicability to fixed-word length EDPMs
such as Univac, and is useful not so much to indicate the number of words
present in highly variable files but to indicate which of a number of data
elements are present in a given master portion or sub-item. One of the
more or less undesirable features of magnetic tape storage is that formats
must be laid out in a manner similar to punch card fields, allowing for
entries which may not exist in some of the items. For example, in a
central stock control file, it may be desired to include for each warehouse
the quantity on hand, the amount on order, the total of back orders and a
summary of sales for the past three months--four data elements in addition to
the warehouse identification. For many items, there will be only one or two
non-zero entries at a number of warehouses, yet each sub-item must be filled
out with zeros to bring it to the constant size. An alternative approach
is to establish a key-word consisting of nothing but "1 If and .to" entries,
each position corresponding to a given data element in the item or subitem. A "1" indicates that the corresponding element is present in the
record; a "0" indicates that it is absent. Zero words are not included in
the i tem--only the significant entries. The sequence of "15" andt'Gs tt in
the key word indicates which ones are included" For two reasons this sort
of technique has not gained much headway at the present time: (1) In character-coded machines such as Univac, only 12 data elements can be keyed in
one word, and many words might be required to key an extremely long record;
and (2) the amount of programming necessary to analyze the key words and
manufacture addresses for desired data elements can get excessive unless
special address-modification and word shifting instructions are provided.
In an EDPM which is binary internally, this technique has definite possibilities for some types of files; the key words can be very few in number
because each binary bit can signify an entire word and binary machines
usually have bit-shifting and conditional transfer instructions so far
not included in straight business EDPMs. Some organizations having IBJl
704s are contemplating the use of this technique in handling payroll jobs
on the 704.
8420
8420
WORD LENGTH IN ELECTRONIC DATA PROCESSING MACHINES
The "words," or basic units of information, comprising business records
are highly variable in length or number of characters. They range all
the way from a single character, usually information in coded form, to
several score -- a part description may require two or three hundred
characters. An analysis of a large number of business records indicates
that about two-thirds of all "words" are six or fewer characters in
length and most of the rest do not exceed twelve characters. As a
further note, "words" greater than this latter length tend to be mostly
alphabetic or alphanumeriC., such as names, addresses and noun descriptions.
Strictly numeric information seldom exceeds ten or eleven digits.
In order to be an efficient business data processor, an EDPM must be capable of performing operations on these variable-length units of information. There are several methods of doing this and the choice of alternate
approaches has given rise, among other things, to the present controversial statements and assertions on the relative merits of "fixed" versus
"variable" word length in an EDPMo The purpose of this section is to
analyze some of the various logics which have been developed to permit
the handling of variable length business t'words tI
0
Most of the electronic computers and data processors which have been
built or announced are "fixed" word length machines, within the meaning
of the definition given in 8421. The major ttvariable·· word length
machines which have been developed are the IBM 702 and 705 and the RCA
BIZMAC. The different logics of these two major equipments (the 705 being little more than a slightly modified 702) involve only two of a
number of possible techniques which could be developed to handle 1tvariable" words. To limit the scope of this discussion" most of the comments
will be based upon the specific methods used in these two machines to
handle tlvariable length" words, leaving out alternative methods which have
not yet been incorporated into announced EDPMS.
Sections 8421 and 8422 are concerned with more or less detailed aspects of
the subjecto The essential sales and procedural implications of fixed vs.
variable word length are contained in 8423, and a knowledge of the other
two sections is not necessary to make use of the information contained in
8423.
Original 4/15/57
8421
8421
DEFINITION OF TERMS
Further discussion will be more explicit with a precise definition of
what is meant by "fixed" and "variable" word length. It should be noted
that the terms are generally used as describing characteristics or attributes of an EDPM; instead of referring to "fixed word length''' as a
noun, it is customarily an adjectival modifier, as in "fixed word length
machine." The definitions are predicated upon this latter usage.
8421.1
Basis for Classificationo The distinction between a "fixed" and "variable'" word length EDPM is based upon the number of bi ts operated upon by
an instru~tiono This can be expressed in another way: The difference
arises from the number of memory accesses for operands required to execute one machine instructiono The term "memory access" means the operation of setting up the internal switching circuitry to make one basic
transfer of information (binary bits) to or from memoryo In most EDPMs,
such an access requires one basic machine cycle of time; in general, the
carrying out of one instruction takes an integral number of these machine
cycles. The definitions following will modify these statements slightly,
but the concept is still valido
8421.2
Fixed Word Length Machineo A fixed word length EDPM is one which performs its operations on a constant and fixed number of binary digits,
called a word. Stated another way, all instructions requiring transfer
of information to and from memory always involve a fixed number of bits;
the exact number is a part of the built-in logic of each EDPM and cannot
be varied by the user. In general, the memory access for one word takes
one basic machine cycle of timeo The tvwordt'l may be a pure binary number,
as in most scientific computers, or a specific number of binary-coded
characters, as in most business EDPMs. The machine itself does not know
the difference and simply operates upon a known and constant number of
bits.
Fixed word length EDPMs may be further subdivided into three categories,
depending upon the method of maki.ng memory accesses. The distinction is
in many respects technical, but a knowledge of the classes facilitates
understanding the reason behind the fixed word length concept.
8421.21
Parallel Word Transfers. Some EDPMs access the entire word in parallel;
that is, the switching circuitry has enough "lines" to provide one for
each bit of the word and all. bits are moved simultaneously. Typical
examples are the Univac Scientific (1103A) and the IBM Types 701-704-709.
All have a 36-bit binary word an~ make one access in one basic machine
cycle for the entire 36 bits" To be of this type, a machine must have a
memory of a form permitting simultaneous access to a complete word; for
practical purposes, this is limited at the present time to Magnetic core
and the obsolescent electrostatic storage deviceso
8421.22
Parallel-Serial Word Transfers. In this type EDPM, several bits are
accessed in parallel, but more than one such access is required to obtain or deliver an entire word" All magnetic drum processors are of this
type; here the bits of each character are handled in parallel but the
successive characters follow one another in sequence. The Univac File
Computer and IBM 650 are in this category; the former moves twelve
Original 4/15/57
8421.22
8421.22 (Continued)
characters in sequence (serially), handling the bits of each character
in parallel, while the latter does the same thing for eleven character
codes (ten digits plus sign). In scientific drum computers, the internal language may be binary, in which Case the bits moved in parallel
are some integral sub-multiple of the word length of the complete binary
word. In magnetic drum computers, the term "word time" is used to indicate the basic overall time required to "access" one complete word.
Magnetic drum EDPMs are of the parallel-serial type as a matter of
economy F~ther than technical necessity. They could easily be modified
to handle an entire word in parallel, but the additional cost of the
switching circuitry would add appreciably to the cost of the equipment.
Likewise, increasing the speed of accesses by a factor of ten or twelve
would be useful only if accompanied by faster operating circuits, which
again cost much more than circuits geared to the speed of character-ata-time access.
8421.23
Serial Word Transferso In a few EDPMs, the transfer of the bits comprIsIng a word is completely serial--that is, one bit at a time. This
type of transfer is typical of EDPMs having a so-called "circulating"
memory, in which the bits are available in a serial fashion only. The
outstanding example is Univac I; its mercury delay-line memory is a
serial type storage device, with the bits available one after another
as they go through the amplifying circuitry before being reintroduced
into the ~ercury tanks. In general, EDPMs of this type have an acoustic
delay line of some form as the high-speed memoryo
8421.3
Variable Word Length Machineo A variable word length EDPM is one which
performs its operations on a varying number of binary digits, rather than
a fixed number. Each character code of an operand requires a separate
memory access, each one requiring one ba~ic machine cycie of time. The
number of accesses required to handle a complete "word" varies with the
number of characters to be handled; because this is usually indeterminate from the instruction itself, special provisions are incorporated
into the machine logic to signify the end of a piece of information.
Although there are numerous methods of accomplishing handling of variable number$ of characters, only the two equipments which have the
feature are discussedo
8421.31
IBM 702/705 Version. It might be more correct to refer to these equipments as having "variable field length" rather than "variable word
length." IBM's variable word length concept, as exemplified in the 702705, is identical with the standard meaning of variable fields in punch
card methodologyo Once a field size for a given word has been established, that word is always handled in the predetermined length.
The 702 is strictly serial-by-character in,all memory accesses. The 705
does have a five-character fixed word length transfer for access to instructions and for the high-speed memory transfer; in both instance, the
full 35 bits of five characters are obtained and moved in p:arallel, using
the same switching circuitry. With these two exceptions (the first of
Original 4/15/57
8421.31
8421.31 (Continued)
which is beyond control of the programmer), the 705 is a purely serialby-character processor.
8421.32 BIZMAC Version. The BIZMAC performs its operations on a character-bycharacter basis, similar to the 702/705, with one significant difference:
Under certain conditions, an operation can be stopped as soon as all
significant characters of a word have been operated upon. As an example,
a seven-digit numeric field might contain, in a record, an entry "0000012,11
which is to be added to something. Only the last two digits are significant and BIZMAC can stop an operation as soon as they have been processed. This is variable word length processing taken one step farther
than the IBM equipments. The use of this feature requires that nonsignificant zeros and blanks in variable length fields be changed to
"item separatorsn-special character codes having no other meaning. The
BIZMAC makes successive memory accesses, one character at a tim~ until
an item separator is encountered; this stops the operation regardless of
the field size. If no separators exist, the specified number of characters are processed. Item separators must be inserted by programmed instructions; they are not generated automatically as a part of machine
operations.
8421.4
Instruction Execution Timeso In a fixed-word length EDPM, each instruction requires a specific ti~e for execution; this time never varies because the operation is always performed on an identical number of characters. (There are, of course, some instructions, such as shifts, which
require varying execution times)o The execution time is almost always an
integral number of basic machine cycles; different types of instructions
require different numbers of cycles and similar instructions in two different computers may differ in the number of cycles required. For instance, in Univac I and II, which have similar logics, the transfer of
one word from memory to a working register requires three basic cycles
(excluding latency time in Univac I): (1) One cycle to obtain the instructions; (2) one cycle to interpret the instruction and set up the circuitry to perform it; and (3) one cycle to obtain the operand from memory
and place it in the register. Some EDPMs do this in two cycles, the
first two steps being combined into one.
A variable word length EDPM has execution times dependent upon the number
of characters involved in an operand. The total time is the sum of a
constant time necessary to obtain the instruction, interpret it and set
up the circuitry (usually two or three basic cycles), plus one cycle for
each character operated upon~
Original 4/15/57
8422
8422
SIGNIFICANCE OF WORD LENGTH IN AN EDPM
From the user's standpoint, whether an EDPM is of fixed or variable word
length in internal operations is of purely academic interest. He is--or
should be--interested primarily in the amount of productive work per dollar which the processor can deliver. To the extent that the word length
in an EDPM affects total processing time and cost, it is a question of
direct interest to him. In this section an attempt is made to outline
the principal implications of fixed word length machines and the 702/705
and BIZMAC versions of variable word length.
8422.1
The User's Point of View. The EDPM user is interested in the equipment
solely as a tool capable of doing productive data processing. His yardsticks are cost and time of doing accurate and useful work. The factors
allocated to cost and time include, obviously, the total charges of the
equipment and its operators, but also encompass "set-up charges"--in the
case of EDPMs, programming and coding.
The performance of productive work by an EDPM involves the time spent in
internal data handling and necessary machine housekeeping-such things as
additions, subtractions, comparisons, data transfers, instruction modifications, etc. In addition, it embraces the time required to move information into the processor to be worked upon and that necessary to
transfer completed work into output storage devices; in today's data
processors, this is tantamount to saying "tape read and write time."
The effect of word length insofar as it affects the ability of the EDPM
to do productive work is of legitimate interest to the user. Academic
discussions of abstract theory belong in research, not in a computer installation.
8422.2
Implications of a Fixed Wor~gth EDPM. In this section are listed a
number of statements of fact and· certain operational considerations pertaining to data processing on a fixed word length machine; these are
basically uncontroversial.
1.
The word length should be long enough to contain the large majority
of business data elements; it is desirable to minimize the number
of double-word (or "double-precision") instruction sequences. Word
lengths of 10-12 digits or characters, and particularly the latter,
meet this requirement.
2.
With a given number of digits assigned to the address portion of an
instruction, the EDPM memory holds several times as many characters
as the maximum possible address. For example, the 12-character word
and 4-digit instruction address of Univac II make possible a total
of 10,000 words, or 120,000 characters.
3.
The working registers of the EDPM--such as Univac's ~A, rX, rF and
rL--can all be of one-word capacity; this is possible because the
maximum number of characters which must be operated upon by anyone
instruction is pre-determined and constant.
4.
The most efficient use of memory requires that short data elements be
combined, several to one word. Otherwise, there is inefficient use
Original 4/15/57
8422.2
8422.2 (Continued)
of the space available for storage of information to be processed.
5.
The most efficient use of input-output facilities also requires that
short data elements be combined into a word. Unless this is done, a
significant percentage of the information on magnetic tape may be
"padding," contributing nothing to production and decreasing the effective rate at which information can be read into and written out
from the processor. Ideally, the information placed on tape should
consist entirely of useful information with no padding.
6.
The combining of several data elements into one word necessitates
"extracting" specific ones when they are required in the processing.
This means that "extract patterns" must be stored in memory, using
some of the available storage capacity, and instructions must be
placed in programs to move the extractors to working registers. The
latter requirement takes some memory space for instructions and some
computer time for their execution.
7.
A single data element contained in two different records may not be
located in corresponding character positions of the two words containing them. In this case, shifting of one is necessary before such
operations as comparison or addition can occur. The shift instructions take memory space and computer time.
8.
Because of the existence of the possibility just mentioned, it is
desirable to place data elements which must be added or compared
in corresponding character positions of their respective words. This
requirement introduces another element which must be taken into account in the establishment of file formats.
9.
Because arithmetic operations are always performed on a fixed number
of digits, the handling of the sign poses no problems. As a matter
of fact, it can be assumed to be positive unless a specific "minus u
character code is included.
This list is not exhaustive, but includes some of the more important aspects applicable to a fixed word length EDPM.
8422.3
Implications of the 702/705 Variable Word Length Method. In this section
are listed a number of statements of fact and certain operational considerations applicable to the specific method chosen by IBM to obtain
variable word length on these two equipmentso Like the foregoing, it is
not exhaustive.
1.
With a given number of digits assigned to the address portion of an
instruction, the EDPM can conta.in only as many characters as can be
expressed by that number of digits. With the 4-digit instruction address of these equipments, IBM has overcome the 10,000-character
limitation only by resorting to use of the zone codes in one digit
of the address to achieve an increase to 40,000.
2.
Because there is no indication in the instruction of the number of
characters to be operated upon, special provisions are necessary to
signify when an operation has been completed.
Original 4/15/57
8422.3
8422.3 (Continued)
a.
Arithmetic operations are terminated by a non-numeric character
at the left of the field involved. Because it is not always possible to locate the number in a position which has an invariably
non-numeric data character in the required place, it may be
necessary to put an otherwise unnecessary character (usually a
blank) in the posi tiono· The various working registers automatically adjust themselves to the size of the operands.
b.
The transfer of non-numeric information between memory and the
working registers is terminated by a special character code "setting" the register size. This requires either that special instructions adjusting the register to the specified size be included in programs, or that one of the auxiliary storage registers
be permanently set aside for a specific data element length. The
method chosen by IBM also requires that different instructions be
used to move non-numeric data than are used for those involving
arithmetic operationso
3.
Because the number of characters to be operated upon varies, registers
in the 705 are much larger than those in a fixed-word length machine~
It contains one of 256 positions, one of 32 and 14 of 16. It is
necessary for instructions to specify which register is involved; the
means of accomplishing this uses the zone code positions of two digits
in the address part of the instruction.
4.
In the 702/705, arithmetic operations in the registers strip the zone
bits from the character codeso Because the address portion of the
instruction uses the zone code bits in three of the four digit positions, addresses can be modified only by the "add to memory" instruction. Comparison of address limits for exits from loops requires constants equal in both the numeric and zone portions to the particular
address being tested.
5.
In EDPMs, arithmetic operations are performed in the same method used
by human beings--from the units position to the left (This is done to
simplify the handling of "carry" digits). Because the 702/705 secure
only one digit at a time from memory, it is then necessary that the
sign be obtained with or before the units digit. IBM has chosen to
indicate the sign with the zone code bits of the units digit. In effect, the units digit is identical with an alphabetic character: A
through I for a plus sign, J through R for a minus. Instructions are
necessary to be sure that the units position is "signed,ft and more
are required to remove the sign when final output information is being prepared. These instructions require memory space and computing
time.
6.
All memory accesses in the 702 are character-by-character, including
those for instructions. As a result, the machine is quite slow, despite a 23 microsecond cycle time. The 705 was modified to provide
parallel access to instructions; the same circuitry is used for the
so-called tlhigh-speed memory transfer," but for nothing else.
Original 4/15/57
8422.3
(2)
8422~3
8422.4
(Continued)
7.
Because the units or addressed position of instructions involving
memory accesses is always matched up with the "units" position of
the specified working register, no attention need be paid to the
relative locations of two corresponding data elements in two different files o
80
Because the word length is variable, within the limits of maximum
register size, only one instruction is required to perform an operation on any number of characters. "Double-precision't operations
required in fixed word length machines are avoided, saving some instructions and, possiblY, some computer time o
90
Although not specifically a part of the variable word concept, the
restrictions surrounding the use of the "high-speed memory transfer"-which is an almost mandatory method of moving data inside the 705-must be considered in setting up file formats. One effect is to increase the number of non-significant "blanks" terminating operands
involved in arithmetic instructions o
Implications of BIZMAC Varigble Word Length Method. Al though the BIZMAC
is subject to some of the comments applicable to the 702/705, the basic
use of variable word length is quite different. In this equipment, not
only can field sizes vary in length, but different elements within one
field can be handled to eliminate non-essential zeros and blankso The
principal implication of the method of using ~item separitd±~tt is the f~ct
that they are not generated automatically, but must be program-inserted.
In addition, even though two operands in an arithmetic instruction may
have separators to terminate the execution cycles, no separator is placed
in the answer; it also must be programmed.
Internally, the use of variable length words in BIZMAC requires a rather
considerable amount of programmingo· There is some question as to whether
the technique was adopted for internal reasons; the separators are used
to "compress" information stored on magnetic tape
That is, only the significant digits or characters plus one separator of each variable-length
word are written on or read from tape" The method adopted does introduce
some progr.amrning to place succeeding elements read into memory locations
which are known.
0
Original 4/15/57
8423
8423
COMPARISON OF FIXED AND VARIABLE WORD LENGTH EDPMs
From the com~ents of 8422, it is evident that both fixed and variable
word length EDPMs have both advantages and disadvantages. Where the
fixed word Univac requires extractors and instructions to use them, the
702/705 "signing" of arithmetic operands or setting of register lengths
and the instructions to do these things, and the BIZMAC needs item separators with the instructions necessary to place them. Where Univac results in some "padding" to fill out occasional words or to expand an item
into a whole number of words, the 705 introduces "paddingn to terminate
arithmetic fields or to fill items out to a multiple of five characters
(to use the "high-speed memory transfer n). Any comparison of the relative merits of fixed versus variable word length, as exemplified by these
equipments, must consider not only the time required to execute anyone
instruction, but also the time required to get ready to execute it. It
must go even beyond this; the effect on input-output time is of definite
importance.
An impartial evaluation of the relative merits of fixed word length, as
used in Univac II, and variable word length, as in th 705, is literally
impossible to make. There are too many ramifications which almost inevitably enter into consideration and are often only partly related to
the word itself. Similarly, the application under discussion is important; completely different conclusions can be derived simply because of
differences in data element sizes, processing requirements, the number
of inputs, and the number and type of outputs. As a typical case of the
seemingly remote factors which must be taken into account, consider this
one simple facet: Univac's fixed word length would be of vastly different significance if the card-to-tape converter and the high speed printer
did not have plugboards. With them, several short data elements in one
word present no particular problem; control panel wiring can scramble or
unscramble them as desired~ With them, internal editing often consists
ofnothing more than a pair of V-W instructions moving several adjacent
words, each of which may have two or three data elements; without them,
more internal instructions would be required to facilitate data edition.
This is precisely what occurs in the variable word length 705, which does
not have a printer plugboard. Yet, seemingly, control panels have no
direct relationship to how Univac or the 705 operate internally.
What can be attacked--and highly successfully -- is the approach that
variable word length is an end unto itself - a sort of "sacred cow" of
data processing methodology. Yet precisely this concept is all too often
encountered. What also can be attacked is the implication that only the
variable word length machine has the "miraculous" ability to handle data
elements of variable sizes. Univac, with its fixed word, is fully as
capable as the 705 of handling variable length data elements; the mere
fact that Univac does it in a different manner and uses different instructions does not mean that it lacks the ability. One simple fact is
often overlooked -- The logic of a fixed word length EDPM is such that
it operates upon all the characters of a word, or any part thereof, in
exactly the same time. As long as it has the ability to isolate a part
of a word from the remainder, it can handle variable length data elements.
Virtually all fixed word length EDPMs have this ability through "extract"
or "buffing" features.
Original 4/15/57
8423
(2)
8423
(Continued)
In a similar vein, neither is fixed word length an end unto itself. It
is simply one technique--and a highly efficient one--of internal EDPM
operations. The fact that these are done on a fixed number of characters
does not mean that data elements must be of that fixed sizeQ After all,
a data processing machine intended for business applications would be
rather ill-designed if it assumed they were.
At this point, the question may be raised; "Yes, the statements you
have made are all well and good. But they are--and I think you will
agree--rather general. What I want to know is this; What can 1 do to
show ~ 'brainwashed' prospect that variable word length is not sacrosanct?" Now this is a perfectly legitimate queryo It must be conceded
that specifics are easier to sell than generalities. Are there any of
these "specifics?" In the vernacular of the day, "You betcha."
Before considering some of these detail points, let's iterate one thing.
We are selling a data processing system--and a darned good one. We are
not selling gither fixed or'variable word length, and they are important
only insofar as they affect how "darned good" our products are. This
point should always be kept uppermost in mind. Now for some brass tacks.
8423.1
What Eguipment Is the Competition Proposing? Our principal comp~tttib.n'
in the EDPM field is coming from International Business Machines Corporation. IBM also happens to be the source of most of the claims for the:>
desirability of variable word length. And IBM also happens to make fixed
word length machines and to propose them as efficient data processors.
Here are the three major classes of equipments IBM is proposing for data
processing, with approximate monthly rental ranges:
Type 650
Type 705
Type 709
$
5,000 - $14,000 monthly
24,000 - 40,000
"
50,000 and up
"
Fixed Word Length
Variable Word Length
Fixed Word Length
Needless to say, IBM claims all of them are the most efficient data processors there are in the respective price ranges.
This simple three-line listing leads to an obvious conclusion: IBM
preaches the ttadvantages" of variable word length only if the prospect
is a potential 705 user. Their salesmen are not even going to mention
the word "variable" if a potential 650 or 709 sale is impending. To be
blunt, IBM talks "variable word length" only to 705 prospects and only
because the 705 happens to have it. This is, of course, a common and
accepted sales approach--seize upon something you have that the competition doesn't and stress its importance.
The use to which this can be put varies with potential customers; the
Univac salesman is the best judge of how to turn this fact of the IBM
line to our advantage. A few possibilities are listed.
To a 705 prospect--which means our proposal involves a Univac II--it may
be apropos to point out IBM's own inconsistencies and apparent uncertainties. If variable word length is so awfully attractive, why don't the
Original 4/15/57
8423.1
8423.1 (Contip~)
650 and 709 have this feature? If variable word length makes the 705
a better data processor, why would the same thing not hold true with
the others?
Perhaps IBM may claim that the volume of work to be performed on a 705
makes variable word length more desirable. There are least two good rebuttals to this. For one thing, small volume applications often require
Just as many different types of operations as the large onesJ there are
as many types of calculations in a 1,000-man payroll and labor cost
distribution as in a 50,000-man one. The only difference is that the
latter has 50 times as many repetitions. If variable word length permits a job to be done faster on the 705, the same thing would hold true
for the 650 0 Secondly, if variable word length is desirable to permit
more work to go through the 705, it would be even more desirable on the
709, which supposedly can handle even more volume.
Or IBM may observe that the logics of the three EDPMs are different, and
only the 705 can profit from variable word length. The obvious retort:
t'Why are they different?" Data processing is data processing and many
typical business fUnctions invlove pretty much the same number of different operations, regardless of volume. Or does the volume of work change
the logic of a machine? If so, why is the small machine fixed word, the
medium variable , and the large fixed?
8423.2
Internal Processing of Data Elements. The advocates of variable word
length EDPMs lay heavy stress on the fact that this feature permits
short data elements to be handled more rapidly than long ones. Why take
the time to handle twelve characters when the element only has two?
A little later, some general remarks will be made about the veracity of
this statement. Right now, consider a few specific factso
8423.21
Processing Frequency of Various Sized Data Elements. About two-thirds
of business data elements are six or fewer characters in length. This is
generally true, but a more pertinent point is this: What is the frequency
of processing data elements of varying lengths? It may be perfectly true
that a given business item has 80% of its data elements three or fewer
characters long. To conclude from this statement that 80}6 of the work
done involves elements of three or fewer characters is a totally unwarranted assumption. 90% of the work may be done on elements of ten or
twelve characters. If this is the case, the "attractiveness" of variable
word length largely vanishes. If there is any advantage to variable word
length machines, it arises only when the average item processed is fairly
short.
A number of frequency studies have been made on the number of characters
in operands from typical Univac I programs. Unfortunately, these are all,
so far as known, static counts;. a program is taken and each instruction
is entered as a spread-sheet tally according to the number of characters
in the data element it is working on. Much more pertinent, and necessary,
is a .<;Iynamic count; tha t is t - a breakdown of the number of characters
operated upon cou~ted once for each time an instruction is executed in a
Original 4/15/57
8423.21
8423.21
(Continued)
machine run. The fact that a~ exception routine has 22 instructions
operating upon elements of one or two characters is not particularly significant if that routine is used only twice in a 50,000-item master file
process., As a further note, some of the static counts include tlhousekeeping tt instructions--such as those to modify memory addresses--and thus
do not afford a reliable picture of even the static frequency of instructions operating upon datao And, finally, counts of the lengths of data
elements made from Univac--or 705 or File Computer or 650--programs are
subject to one other restriction which may partlally invalidate their
significance: To some extent, the "field size" of a data element is subject to numerous considerations other than central computer processing,
and may be shorter or longer than necessary to facilitate periphery operations.,
Although the factors outlined above restrict our detailed knowledge of
the effects of variable data elements, some specific comments can, nonetheless, be made.
8423.22
Tape Limited Processing. So long as an EDPM is in a tape-limited process,
the actual computing time required to do whatever is necessary is practically immaterial. If a tape input/output rate of 1,000 characters every
50 milliseconds is the limit that can be achieved with the tape units in
use, it is immaterial whether the processing on that 1,000 characters
takes 45 milliseconds, or 5. There is no gain in processing speed by
increasing the amount of time the main frame is "spinning its wheels."
How much time does the average EDPM spend in this status? Well, on Univac I, about 50%. On Univac II, it is estimated that 65% of the time in
an average installation will be tape-limited.
Now this may be countered by IBM by noting that the 705 is "so fast" internally that it can use two or three Type 777 Tape Record Coordinators
and handle several sets of tapes simultaneously. This is perfectly true.
It can. And it is still slower than Univac II with its one faster set
of input-output buffers and tapes. It also cos ts more. The proo·f of
this is developed in detail in Section 1180-1189.
8423.23
Computer Limited Processing" Although there is no gain by speeding up
computational speed when the machine run is tape limited, it would be an
advantage in the computer-limited runs. The claim is that the 705 can
do this with its variable word length instruction execution. In some
cases, this quite possibly is the case. However, let's consider some
of the typical cases which are computer limited.
Some sort-merge operations are in this class, particularly with fairly
short items.. Aside from the tape-handling sequences, this type of run
involves three basic phases: (1) Comparison of index numbers; (2) modification of addresses; and (3) internal transfers of items. The last
has nothing to do with the length of words. How about the first two?
Index numbers tend to be longer than "average" data elements; they are
not often less than six characters, and such common indices as names,
part numbers, customer account numbers, social security numbers, job order
Original 4/15/57
8423.23
8423.23 (Continued)
or project number, etc., run anywhere from eight or nine up to 24 or
more characters. And in this range the 705 takes as long as Univac II
to handle an operand.
Consider address modification, which is common in sort-merge routines.
Both Univac II and the 705, for example, have four-digit addresses. The
fastest Univac modification is a 'tB" and "AH" pair of instructions, which
take a total of 360 microseconds. The 705, using a "Load" and "Add to
Memory" pair, takes 204, which is obviously faster. Furthermore, it may
be possible to keep the modifying increment in an auxiliary storage
register and reduce the time to 102 microseconds. The 705, then, is
definitely faster than Univac II by better then three to one. Of course,
this conclusion overlooks the fact that I can modify two addresses in
Univac in the same amount 'of time--and, surprisingly enough, both addresses
of a V-W pair to move an item to its proper place may be involved, depending upon the exact sort-merge technique followed. Furthermore, it
neglects to consider that limit tests on the new address are necessary-and Univac may do this with generalized overflowo The 705 won't--at a
minimum, two more instructions (nCompare" and a conditional transfer),
totalling 136 microseconds, are required. And it may take more if constants can't be contained in one of the auxiliary storage registers.
(one of the baffling aspects of competitive proposals is how IBM assumes
that 15 auxiliary storage registers can hold 60 or 80 constants at one
time).
Another common type of computer-limited process is a payroll and labor
cost distribution;, in terms of processing requirements, such functions
as insurance premium and cash value determination, public utility customer
billing and accounting, and inventory analysis are similar. What are the
sizes of data elements involved? To be sure, some comparisons on one-or
two-character codes, but the bulk of computer time (aside from transfers
into and out of working storage and address modification) is spent on
arithmetic operations: Multiplications of two factors ranging from three
to six or seven digits to arrive at rounded products of from five to nine
or ten digits; testing of answers of six to eight digits against limits
of the same size; accumulating a wide variety of totals ranging from six
to ten or eleven digits in length. Much of the computing time is spent
on medium to fairly long length data elements, not on extremely short
ones. Of course, the fact that Univac can deliver a rounded product as
desired, whereas the 705 cannot, possibly is unrelated to word length; the
fact that it is fairly easy to build in this facility with a fixed word
length machine and a good deal more difficult in one with variable word
length does have adifinite effect on comparative abilities to arrive at
required answers.
Likewise foreign to fixed versus variable word length is the fact that
Univac gives correct answers because of its self-checking circuitry.
It is understood that some auditors have acquired the habit of insisting that accounting procedures provide for accuracy; although this requirement likewise is unrelated to an EDPMs word length, it has an adverse effect on the 705's ability to deliver reliable results.
Original 4/15/57
8423,,3
8423.3
Relationship of Word Length to Item Length" A business item (record, in
IBM terminology) consists of a number of data elements of varying lengths
and items within one file typically are of variable lengtho In the establishment of file formats it is necessary to consider not only the
method by which the processor handles variable length data elements, but
also the means available to transfer information internally to reassemble
a new master file, taking into account expansions and contractions, or
to reassemble various pieces of several files into a new output--say,
one for printingo
In the 705, IBM has utilized one version of a variable data element or
word length machine--and at the same time made it somewhat awkward to use
with variable length items or records. The 702 was strictly a variable
word length machine--every instruction, every operand and every transfer
was obtained and moved character by charactero To be more precise, the
702 might be called a machine with a fixed word length of one character,
with instructions operating on successive "words" until something stops
the cycle" With the 705, IBM recognized that this character by character
access--in other terms, their "variable word length"--was not always a
desirable attributeo For example, every EDPM instruction must be obtained
from memory and, because it is always of fixed length, why not obtain it
in one memory access instead of several? So they did--put in a fivecharacter access line" By doing this, IBM admitted that variable word
-length was no panaceao But they went even further--they put in hardware
to use this same five-character access line to speed up internal memory
transferse
Well, the choice made gears memory transfer to a record or item, and
imposes restrictions on the item formate The basic difficulty is that
the use of the five-word memory transfer, when variable length items are
being handled, involves a lot of address modification and associated
limit tests, all of which take instructions and time" Moreover, it canit
always be used, particularly in assembling bits and pieces of several
items into a new one, necessitating slow character-by-character movement.
(This is discussed in detail in the sections on the Type 777 Tape Record
Coordinator, 1180, and the 705 central processor, 5140)0 They have gone
even farther, input from and output to tape is also geared to records
and is most efficient when all tape blocks are the same sizeo In effect,
they have built an equipment which operates internally on variable
length words, in part, but is used most efficiently with fixed length
items"
Now consider Univac.. Its internal operations on words, its method of
making memory transfers and its systems of storing information on tapes
are all completely independent of data element or item sizee It handles
fixed or variable words and fixed or variable items with complete impartiality and flexibilityo
Now the fact that an EDPM operates internally on a variable word length
basis may seem only remotely connected to memory transfers or storage
of information on tape" In theory, this may be true, but we are selling
hardware against hardware, not theory~ In attempting to overcome certain
weaknesses in a completely variable word length machine, IBM has limited
the effectiveness of the 705 internal memory transfer and seriously restricted flexibility, and increased the processing time, of magnetic tape
files"
Original 4/15/57
8423.4
8423.4
Does Variable Word Length Provide Faster Internal Processing? The proponents of variable word length claim that processing is faster because
instructions operate only upon the specific number of characters in a
data element. It is obvious, they claim, that a comparison on a onecharacter element is faster if the equipment looks only at that character
than if it handles a full word of, say, 12 characters. How valid is this
claim?
Before considering the facts, a precautionary word should be noted. It
is unfortunate that, throughout the EDPM field, comparisons between two
or more equipments are made on the superficial and totally inadequate
basis of a single instruction execution time. As always, the true measure of a data processor's efficiency is its cost of doing a certain
amount of productive work. OneEDPM, for example, may be able to execute
individual instructions twice as fast as another. But if it takes three
times as many instruction executions to perform a given process, it rather
obviously is not the faster machine. The types of instructions provided are at least equally important with the speed of their individual
executions.
Although the generality of the conclusions is subject to restrictions for
these reasons, it is possible to analyze, at least in part, the validity
of this claimo For this purpose, let us consider existing equipments
with known characteristics, rather th.an attributes which could be included in an electronic data processor, but haven't been.
The equipments to be used are Univac II and the IBM 704 and 705, all of
which are single-addressed EDPMs. Of these, only the 705 has the so-called
"variable word length.;" Let us look at the times required to execute
the addition instruction; in all three equipments, this instruction obtains an operand from memory, adds it to the contents of a register and
places the sum in the same registero The following tabular presentation,
with times in microseconds, is appli£able:
Factor
Word Length (characters)
Basic Cycle Time
Cycles to Perform an Addition
Univac II
12
40
5
36 bits
12
2
Variable
17
2/ N
Some interesting conclusions can be derived. The logics of the equipments
differ in terms of the basic cycles required to complete the addition.
The 704, which has the fastest cycle, requires only two of them to add
two 35-bit plus sign numbers; this is equivalent to ten or eleven decimal
digits. The 705, which has a basic cycle about 45% longer, can add two
one-digit numbers in 51 microseconds--more than twice the time required
by the 704. It takes 204 microseconds to add two 10-digit numbers--700%
longer than the 704, although its cycle rate is only 45% longer~
Now compare Univac II and the 7050 UnivacQs cycle rate is 135% longer~
And quaintly enough, Univac's addition time for a three-digit number is
almost exactly 135% more than that of the 705. And this leads directly
to a pertinent observation: In adding anything oyer three-digit numbers,
the 705's faster execution time is due not to its variable word length
Original 4/15/57
8423.4
(2)
8423.4
(Continued)
but to its faster cycle timeo Or take Univac's "Bro" instruction, which
loads one word into rA in 120 microseconds or three basic cycles, with
the 705' s "Load,t order, which moves one character into an accuinula tor
in 51 microseconds. and succeeding characters in 17 microseconds more
each. Univac's time is almost exactly 135% more than the 705's time
for one character; in other words, for many comparable instructions, ~
apparent speed advantage of the 705 over Univac is due solely to its
faster cycle rate, regardless of the number of characters moved. And it
might be added that the 705's actual time is greater if more than five
characters are involved.
A number of studies have been made of the comparative speeds of Univac
II and the 705 in executing actual operating instructions. These indicate that, on the avera~e and instruction for instruction, the 705 can
go through about 2t - 22 instructions to one for Univac. It may only be
coincidence that Univacvs cycle time is about 2 1/3 times that of the 705.
Whether it is or not, one conclusion is inescapable: The fact that the
705 has a faster instantaneous instruction execution rate than Univac II
is traceable almost exclusively to its faster cycle time; the variable
word length feature is only a contributing factor of must lesser import.s.n.£.§..
The statement in the previous paragraph of the relative instruction execution rates of Univac II and the 705 should not be misconstrued; it is
based on individual instructionso Interpreting it to mean that the 705
can do 2t - 2~ times as much work as Univac II is completely unjustified.
And fallacious.
8423.5
Programming Ease. In theory, there should be no difference in the programming ease or difficulty of a fixed as compared with a variable word
length EDPM. In actuality, the particular gimmicks incorporated in both
the 702/705 and BIZMAC make them more difficult to program than Univac.
With the 705 particularly, the incidence of clerical errors in coding appears to be much higher than that of our own equipments. This is not
Remington Rand Univac opiniono It has been stated by several independent,
and impartial, management consultants who have had occasion to analyze
the operations of the two equipments~ The section on the 705 central
processor discusses this aspect in more detailo
8423.6
What Is the Future of Variable Word Length Machines? The remarks of
842304, coupled with the fact that IBM--as the major exponent of variable
word length--is proposing EDPMS of both types for business data handling,
naturally give rise to the question: "What type of word length will the
EDPM of the future have?" This is a perfectly valid request for a customer to make; it is indeed somewhat confusing for one manufacturer to
extol the virtues of both types, and to do it simultaneously.
Any attempt to forecast the future is risky~ It is particularly hazardous
in a fied which has been progressing as rapidly--both technically and
operationally--as that of electronic data processing equipments. The development and perfection of new techniques, availability of improved components and circuitry, perfection of faster hardware and the lowering of
Original 4/15/57
8423.6
8423.6 (Continued)
production costs of components and circuits are of such frequent occurrence as to be prosaico From the product planning aspect, the knowledge
and experience gained from existing computers, coupled with technical
developments, resulting in system concepts unthought of or impracticable
only a short time agoo Although it cannot be expected that the rate of
growth and improvement will continue indefinitely at the pace of the last
few years, there is no present indication of an immediate slowdowno And
so, because sudden breakthroughs on the research front--both technical
and logical--may open up completely new avenues for exploration, a look
to the future--even the relatively near future--becomes pretty much a
"crystal ball" processo
But for what it's worth, one person's thoughts for the very shorth term-say three or four years--followso The 705 and BIZMAC are the last of the
equipments which will have variable word length of the type they represent;
modified versions may come out, of courseo This statement refers to the
large-scale, milli,on-dollar class of machine and is based upon this line
of reasoning. Magnetic core storage appears to be the principal type
which will be used in this class equipment for the next few yearso Although constant speed and cost improvements are being realized, it still
represents a fairly expensive cost-per-bit storage medium, and the cost
increases rapidly as the "speed" of the cores is increased; here "speed"
means the rapidity with which successive accesses to storage can be madeo
Existing core storage units use what is called ttdestructive read-out;"
that is, the information in the cores is destroyed when it is read and
must be restored or "regenerated" to keep it availableo Consequently,
the memory access time, usually equivalent to a basic machine cycle, is
in two parts: (1) The reading phase, at the conclusion of which the bits
are available in the working registers of the equipment; and (2) the regeneration phaseo Cycle times of 12 microseconds are available today,
and this may be cut'in half, at no additional cost, in the next two or
three years. Today, storage of this speed represents a selling price
of about one dollar per bit, not exactly inexpensiveo
Considerably faster storage devices can be built today, but· the costper-bit is prohibitively expensive if any volume is involvedo They are,
however, not so expensive that a few of them can't be used as the working registerso This is what is done in Univac II, the Univac Scientific
and the IBM 704 and 7090 These processors consist of large core storage
memories of (relatively) slow speed and a few much. faster registers-faster by a factor of possibly ten to twentyo Consequently, it becomes
feasible to develop a computer with instructions executed along these
lines: (1) Obtain the instruction during the first half of the basic
cycle and move it into the high-speed instruction register; (2) in the
next half-cycle, "regenerate" the instruction in memory and simultaneously
decode it in the instruction register and set up the necessary circuitry
for execution; (3) in the next half-cycle, obtain the operand from
memory; and (4) in the next half-cycle, regenerate the operand in memory
and simultaneously process it in the hig~-speed registers. This is the
essential cycle in the Univac Scientific and the IBM 704 and 709. Now
it is evident that so long as two memory accesses are required for the
Original 4/15/57
8423 6
0
(2)
8423.6
(Continued)
instruction and the operand, there is actually a loss of processing speed
if the equipment secures operanda character by character when it is fast
enough to operate on several characters during the necessary regeneration
half-cycle. As long as the present methods of using magnetic core storage
exist, fixed word length operation will be much faster than variable, even
allowing for instruction time which might be necessary to handle extractors for data elements shorter than the word length~
Of course, progress is being made on cheaper methods of assembling magnetic core matrices, on cores which can be operated at higher rates of
speed without a cost increase, and on methods of non-destructive readout,
which will eliminate the need for regeneration. Although the availability
9f these improved mass storage units may change present concepts, it appears that, at least for the immediately foreseeable future, a largescale data processor can economically include a few working registers
capable of operating at speeds several times as great as the memory_ If
this condition prevails, fixed word length operation probably will provide the faster processingo
The foregoing comments may assist in placing the variable word length
702/705 in an historical perspectiveo Their basic logic was settled
several years ago, at a time when IBM was using electrostatic storage as
the high-speed memory. Just why the character-by-character access of the
702 was adopted, when a similar type memory used a full word access in the
701, is not known; it may well have provided an economic technique at the
time. Memory switching requirements are somewhat less and registers can
be slower than a fixed word length machine
Thus it may well have been
that consideration of availabLe techniques and costs of achieving them
were highly important criteria in settling upon the single character mode.
Certainly the fact that IBM is building both variable and fixed word length
processors indicates that there is within the corporation no unanimity of
opinion that variable word length is best for the userQ
0
In the large-scale EDPM fieldS) thenS) onE? opinion is that fixed word length
will be a characteristic of equipments to be announced in the very few
years to come o In the medium and low-price fields, there is possibility
of even more diversity of opinionQ The medium I':{lnge may utilize fairly
slow, but not too expensive, magnetic core 'storage units of small size,
backed up by drums or random access fileso In this case, the internal
word (but not necessarily the storage word) may be fixed in length; this
is by no means certainQ In the lower-priced field and in the medium-cost
equipments using magnetic drums as the main storage medium, fixed word
length prohably can be expectedo Although magnetic drum computers could
be built with a variable length w0rd~ the existence of latency time and
the desire for low selling cost makes any gain from variable word length
memory access quite marginalo Consequently~ one school of thought feels
that variable word length of the type used in the 702/705 is not a particularly attractive possibility in this range of equipmentso
There are obviously other opinionso With progress continuing at impressive
speed, there is nothing more uncertain than attempting to predict what an
EDPM of two or three years from now may incorporateo
Original 4/15/57
842307
8423.7
Conclusion o The foregoing are just some of the possibilities for the not
too distant futureo They mayor may not be indicative of what will happeno A technical development announced tomorrow could easily make completely new concepts possibleo
But, if nothing else, this discussion may have helped to place the variable word length feature of the 702 and 705 in its proper place as only
a part of a data processing equipment, and to consider it as just another
step in the evolution of electronic data processing techniques and logico
There would be little objection if IBM claims for the variable word
length feature of these two equipments were based largely on its being
a technically feasible and, possibly, economical method of computer
operationo Unfortunately, such an approach has little sales appeal.
But there is definite objection to IBM claims that the variable word
length concept of the 702 and 705 provide the best systems approach for
handling variable length business data elementsQ
IBM itself has provided part of the rebuttal by building--and of course
claiming superiority for--fixed word length data processorso Several
other rebuttal points are included within this section. They may be
summed up qui te briefly; The function of an EDPM is to pro.cess business
data. Provided an equipment operating with a fixed length internal word
can handle variable length data elements, there is no reason it cannot be
an effective processor.
Original 4/15/57
Source Exif Data:
File Type : PDF File Type Extension : pdf MIME Type : application/pdf PDF Version : 1.3 Linearized : No XMP Toolkit : Adobe XMP Core 4.2.1-c043 52.372728, 2009/01/18-15:56:37 Create Date : 2010:08:31 20:59:05-08:00 Modify Date : 2010:09:01 09:28:19-07:00 Metadata Date : 2010:09:01 09:28:19-07:00 Producer : Adobe Acrobat 9.33 Paper Capture Plug-in Format : application/pdf Document ID : uuid:77a456b7-bb26-45c3-afa5-2d2b5b866ed6 Instance ID : uuid:b2d8cfe8-6656-46ef-b4f1-6fab451a6f5e Page Layout : SinglePage Page Mode : UseNone Page Count : 282EXIF Metadata provided by EXIF.tools