Btt Manual
btt-manual
User Manual:
Open the PDF directly: View PDF 
.
Page Count: 34
| Download | |
| Open PDF In Browser | View PDF |
btt User Guide Alan D. Brunelle (Alan.Brunelle@hp.com) 30 October 2008 1 Introduction btt is a post-processing tool for the block layer IO tracing tool called blktrace. As noted in its Users Guide, blktrace is a block layer IO tracing mechanism which provides detailed information about request queue operations up to user space. blktrace is capable of producing tremendous amounts of output in the form of multiple individual traces per IO executed during the traced run. It is also capable of producing some general statistics concerning IO rates and the like. btt goes further and produces a variety of overall statistics about each of the individual handling of IOs, and provides data we believe is useful to plot to provide visual comparisons for evaluation. This document will discuss btt usage, provide some sample output, and also show some interesting plots generated from the data provided by the btt utility. A short note on the ordering of this document – the actual command-line usage section occurs relatively late in the document (see section 11), as we felt that discussing some of the capabilities and output formats would make the parameter discussion easier. This document refers to the output formats generated by btt version 2.00. However, the descriptions are general enough to cover output formats prior to that. 1 Contents 1 Introduction 1 2 Getting Started 3 3 Output Overview 4 4 Data Files Output 11 5 Activity Data File 12 6 Histogram Data Files 14 7 iostat Data File 16 8 Per-IO Data File 17 9 Latency Data Files 19 10 Seek Data Files 20 10.1 Seeks Per Second . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 11 Command Line 11.1 --seek-absolute/-a . . . . . . . . . 11.2 --all-data/-A . . . . . . . . . . . . . 11.3 --dump-blocknos/-B . . . . . . . . . 11.4 --range-delta/-d . . . . . . . . . . . 11.5 --devices/-D . . . . . . . . . . . . . . 11.6 --exes/-e . . . . . . . . . . . . . . . . 11.7 --help/-h . . . . . . . . . . . . . . . . 11.8 --input-file/-i . . . . . . . . . . . . 11.9 --iostat/-I . . . . . . . . . . . . . . 11.10--d2c-latencies/-l . . . . . . . . . 11.11--periodic-latencies/-L . . . . . . 11.12--seeks-per-second-m . . . . . . . . 11.13--dev-maps/-M . . . . . . . . . . . . . 11.14--output-file/-o . . . . . . . . . . . 11.15--per-io-dump/-p . . . . . . . . . . . 11.16--per-io-tress/-P . . . . . . . . . . 11.17--q2c-latencies/-q . . . . . . . . . 11.18--active-queue-depth/-Q . . . . . . 11.19--no-remaps/-r . . . . . . . . . . . . 11.20--seeks/-s . . . . . . . . . . . . . . . 11.21--iostat-interval/-S . . . . . . . . 11.22--time-start/-t and --time-end/T . 11.23--unplug-hist/-u . . . . . . . . . . . 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 23 23 24 24 24 24 24 24 25 25 25 25 25 25 25 25 25 26 26 26 26 26 26 11.24--version/-V . . . . . . 11.25--verbose/-v . . . . . . 11.26--easy-parse-avgs/-X 11.27--q2d-latencies/-l . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 27 27 29 12 bno plot.py 30 13 Sample btt Output 32 3 2 Getting Started The simple pipeline to get going with btt is to perform the following steps: 1. Run blktrace, specifying whatever devices and other parameters you want. You must save the traces to disk in this step, btt does not work in live mode. 2. After tracing completes, run blkrawverify, specifying all devices that were traced (or at least on all devices that you will use btt with – section 11.5 shows how you can dictate which devices to use with btt). If blkrawverify finds errors in the trace streams saved, it is best to recapture the data – utilizing btt on unclean trace files produces inconsistent results. While this step is optional, we have found that performing this helps to ensure data coming from btt makes the most sense. 3. Run blkparse with the -d option specifying a file to store the combined binary stream. (e.g.: blkparse -d bp.bin ...). blktrace produces a series of binary files containing parallel trace streams – one file per CPU per device. blkparse provides the ability to combine all the files into one time-ordered stream of traces for all devices. 4. Run btt specifying the file produced by blkparse utilizing the -i option (e.g.: btt -i bp.bin ...). 4 3 Output Overview The major default areas of output provided by btt include: average component times across all IOs The time line of each IO is broken down into 3 major regions: 1. Time needed to insert or merge an incoming IO onto the request queue. This is the average time from when the IO enters the block IO layer (queue trace) until it is inserted (insert trace). This is denoted as Q2I time. This is also broken down into two component times1 : Q2G Time needed to get a request (get request trace). G2I Time needed to put that request onto the request queue (insert trace). For merged requests – an incoming request that is merged with a previously submitted request – we calculate Q2M, the amount of time between the queue trace and the merge trace. 2. Time spent on the request queue. The average time from when the IO is inserted or merged onto the request queue, until it is issued (issue trace) to the lower level driver. Referred to as I2D time2 . 3. Driver and device time – the average time from when the actual IO was issued to the driver until is completed (completion trace) back to the block IO layer. This is referred to as the D2C time Two other sets of results are presented in this section: 1. Q2Q which measures the time between queue traces in the system. This provides some idea as to how quickly IOs are being handed to the block IO layer. 2. Q2C which measures the times for the complete life cycle of IOs during the run3 1 On occasion there are also some time spent sleeping waiting for a request. That occurs between the Q and G operations. You will see these listed as S2G times. 2 The issue trace is represented by a D in the blkparse output, hence its usage in btt to refer to issue traces. Note that an I is used to refer to insert traces. 3 One of the areas that needs some work in btt is to better understand the multiplex nature of IOs during a run. In theory, one would like Q2I + I2D + D2C = Q2C however, typically there are multiple queue traces that are combined via merges into a single IO issued and completed. We currently average the queue-to-insert and queue-to-merge times, and thus tend to be quite close to the expected equation. 5 For each row in this output, we provide a minimum, average, maximum (which are all presented in seconds), and overall count. As an example4 : ALL MIN AVG MAX N ---- ------------- ------------- ------------- ----------Q2Q 0.000000058 0.000012761 9.547941661 2262310 Q2I 0.000000272 0.000005995 0.104588839 2262311 I2D 0.000001446 0.094992714 0.239636864 2262311 D2C 0.000193721 0.030406554 1.634221408 2262311 Q2C 0.000207665 0.125405263 1.830917198 2262311 When tracking device mapper devices, we also break down the Q2A and Q2C times for those IOs. Device Overhead Using the data from the previous chart, we can then provide some idea as to where IO spend most of the time on average. The following output shows the percentage of time spent in each of the phases of an IO5 DEV ---------( 8, 80) ---------Overall | Q2G G2I Q2M I2D D2C | --------- --------- --------- --------- --------| 0.0013% 0.0004% 0.0006% 88.5005% 11.4988% | --------- --------- --------- --------- --------| 0.0003% 0.0001% 0.0041% 21.4998% 78.4958% Device Merge Information A key measurement when making changes in the system (software or hardware) is to understand the block IO layer ends up merging incoming requests into fewer, but larger, IOs to the underlying driver. In this section, we show the number of incoming requests (Q), the number of issued requests (D) and the resultant ratio. We also provide values for the minimum, average and maximum IOs generated. Looking at the following example: DEV | #Q #D Ratio | BLKmin BLKavg BLKmax Total ---------- | ------- ----- ----- | ------ ------ ------ ------( 68, 64) | 2262311 18178 124.5 | 2 124 128 2262382 4 As with this display, the author has taken some liberty in reformatting the output for better display on the printed page. 5 It should be noted that incoming requests either go through: 1. Q2G + Q2I or 2. Q2M before proceeding to I2D and D2C. 6 we see that (on average) the block IO layer is combining upwards of 125 incoming requests into a single request down the IO stack. The resultant average IO size is 124 blocks. Device Seek Information Another useful measure is the variability in the sector distances between consecutively received – queued and submitted – issued IOs. The next two sections provides some rudimentary statistics to gauge the general nature of the sector differences between IOs. Values provided include the number of seeks (number of IOs submitted to lower level drivers), the mean distance between IOs, the median value for all seeks, and the mode - the value(s) and the counts are provided for the latter. The first of the two sections displays values for Q2Q seek distances – providing a set of indicators showing how close incoming IO requests are to each other. The second section shows D2D seek distances – providing a set of indicators showing how close the IO requests are that are handled by underlying drivers. DEV | NSEEKS MEAN MEDIAN | MODE --------- | ------ ------- ------ | ------( 68, 64) | 18178 19611.3 0 | 0(17522) We have almost exclusively seen median and mode values of 0, indicating that seeks tend to have an equal amount of forward and backwards seeks. The larger the count for the mode in comparison to the total number of seeks is indicative as to how many IOs are coming out of the block IO layer in adjacent sectors. (Obviously, the higher this percentage, the better the underlying subsystems can handle them.) Request Queue Plug Information During normal operation, requests queues are plugged and during such times the IO request queue elements are not able to be processed by underlying drivers. The next section shows how often the request queue was in such a state. DEV | # Plugs # Timer Us | % Time Q Plugged --------- | ------- ---------- | ---------------( 68, 64) | 833( 0) | 0.356511895% There are two major reasons why request queues are unplugged, and both are represented in the above table. 1. Explicit unplug request from some subsystem in the kernel. 2. Timed unplugs, due to a request queue exceeding some temporal limit for being plugged. The total number of unplugs is equal to the number of plugs less the ones due to timer unplugs. 7 IOs per Unplug & Unplugs-due-to-timeout In this subsection one can see the average number of IOs on the request queue at the time of an unplug or unplug due to a timeout. The following sample shows a sample of both unplug sections: ==================== Plug Information ==================== DEV ---------( 8, 0) ( 8, 32) ---------Overall Average | # Plugs # Timer Us | ---------- ---------| 1171( 123) | 4( 0) | ---------- ---------| # Plugs # Timer Us | 587( 61) DEV ---------( 8, 0) ( 8, 32) ---------DEV Overall | IOs/Unp | ---------| 9.2 | 2.5 | ---------| IOs/Unp | 9.2 | | | | | | | % Time Q Plugged ---------------0.280946640% 0.000325469% ---------------% Time Q Plugged 0.140636055% IOs/Unp(to) ---------8.8 0.0 ---------IOs/Unp(to) 8.8 This table and the preceding one have to be considered together – in the sample output in the immediately preceding table one can see how the larger number of data values for device (8,0) dominates in the overall average. Active Requests At Q Information An important consideration when analyzing block IO schedulers is to know how many requests the scheduler has to work with. The metric provided in this section details how many requests (on average) were being held by the IO scheduler when an incoming IO request was being handled. To determine this, btt keeps track of how many Q requests came in, and subtracts requests that have been issued (D). Here is a sample output of this sections: ==================== Active Requests At Q Information ==================== DEV ---------( 65, 80) ( 65,240) ... ( 66,112) ---------- | Avg Reqs @ Q | ------------| 12.0 | 16.9 | 44.2 | ------------8 Overall | Avgs Reqs @ Q Average | 17.4 9 Detailed Data In addition to the default sections output, if one supplies the --all-data or -A argument (see section 11.2) to btt further sections are output: Per Process As traces are emitted, they are tagged with the process ID of the currently running thread in the kernel. The process names are also preserved, and mapped to the ID. For each of the parts of the time line discussed above on page 4, a chart is provided which breaks down the traces according to process ID (name). One must be aware, however, that the process ID may not have anything to do with the originating IO. For example, if an application is doing buffered IO, then the actual submitted IOs will most likely come from some page buffer management daemon thread (like pdflush, or kjournald for example). Similarly, completion traces are rarely (if ever?) going to be associated with the process which submitted the IO in the first place. Here is a sample portion of this type of chart, showing Q2Q times per process: Q2Q ------------mkfs.ext3 mount pdflush MIN ----------0.000000778 0.000000885 0.000000790 AVG ----------0.000009074 0.000672513 0.000006752 MAX N ----------- ------1.797176188 1899371 0.030638128 73 0.247231307 179791 Per Process Averages The average columns from the above charts, are also presented in their own chart. Per Device Similar to the per-process display, btt will also break down the various parts of an IOs time line based upon a per-device criteria. Here’s a portion of this area, displayed showing the issued to complete times (D2C). D2C --------( 65, 80) ( 65, 96) ( 65,112) ( 65,128) ( 65,144) ( 65,160) ( 65,176) ( 65,192) ... MIN ----------0.000140488 0.000142762 0.000145221 0.000141896 0.000140832 0.000139915 0.000141254 0.000141453 AVG ----------0.001076906 0.001215221 0.001254966 0.001159596 0.001290985 0.001175554 0.001104870 0.001234460 MAX ----------0.149739869 0.173263182 0.124929936 0.775231052 0.211384698 0.073512063 0.073231310 0.167622507 N -----169112 155488 165726 169015 210661 133973 145764 140618 Per Device Averages The average columns from the above charts, are also presented in their own chart. 10 Q2D Histogram A display of histogram buckets for the Q to D times – basically, from where an IO enters the block IO layer for a given device, and when it is dispatched. The buckets are arranged via the time in seconds, as in: ==================== Q2D Histogram ==================== DEV --------( 66, 80) ( 65,192) ( 65,128) ... ( 65, 64) ( 66, 64) ========== AVG | <.005 <.010 <.025 <.050 <.075 <.100 <.250 <.500 < 1.0 >=1.0 | ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== | 61.2 7.9 12.1 7.9 3.0 1.4 1.5 0.2 0.0 4.6 | 42.3 5.0 8.7 30.0 8.9 3.0 1.8 0.1 0.0 0.1 | 34.3 5.3 8.9 32.0 9.7 3.7 5.3 0.6 0.0 0.1 | 59.9 4.2 6.0 24.6 4.2 0.8 0.1 0.0 0.0 0.1 | 62.6 8.1 12.7 7.9 2.4 0.6 0.1 0.0 0.0 5.4 | ===== ===== ===== ===== ===== ===== ===== ===== ===== ===== | 52.9 6.2 10.0 20.1 5.3 1.7 1.4 0.2 0.0 2.1 11 4 Data Files Output Besides the averages output by default, the following 3 files are also created with data points which may be plotted. file.dat This file provides a notion of activity for the system, devices and processes. The details of this file are provided in section 5. file qhist.dat Provides histogram data for the size of incoming IO requests, for more information see section 6. file dhist.dat Provides histogram data for the size of IO requests submitted to lower layer drivers, for more information see section 6. In addition to the default data files output, there are optional data files which can be generated by btt. These include: subset of .avg data, easily parsed When the -X option is specified and the -o has also been specified, then a subset of the data produced by default is copied to another file that is more easily parsed. Refer to section 11.26 for full details. iostat iostat-like data can be distilled by btt, and is described in section 7. per IO detail Each and every IO traced can be output in a form that shows each of the IO components on consecutive lines (rather than grepping through a blkparse output file for example). The details on this file is included in section 8. iostat Latency information – both Q2d, D2c and Q2C – on a per-IO basis can be generated. These are described in section 9. seek details A set of data files containing all IO-to-IO sector differences can be output, with details found in section 10. unplug histogram details A data file per device containing histogram output for the amount of IOs released at unplug time. Section 11.23 has more details. 12 5 Activity Data File The activity data file contains a series of data values that indicate those periods of time when queue and complete traces are being processed. The values happen to be in a format easily handled by xmgrace6 , but is easy to parse for other plotting and/or analysis programs. The file is split into pairs of sets of data points, where each pair contains a set of queue activity and a set of completion activity. The points are presented with the first column (X values) being the time (in seconds), and the second column (Y values) providing an on/off type of setting. For each pair, the Y values have two settings off (low) and on (high). For example, here is a snippet of a file showing some Q activity: # Total System # Total System : q activity 0.000000000 0.0 0.000000000 0.4 0.000070381 0.4 0.000070381 0.0 1.023482637 0.0 1.023482637 0.4 6.998746618 0.4 6.998746618 0.0 7.103336799 0.0 7.103336799 0.4 17.235419786 0.4 17.235419786 0.0 26.783361447 0.0 26.783361447 0.4 26.832454929 0.4 26.832454929 0.0 28.870431266 0.0 28.870431266 0.4 28.870431266 0.4 28.870431266 0.0 What this indicates is that there was q activity for the system from 0.000000000 through 0.000070381, but was inactive from there to 1.023482637, and so on. Section 11.4 contains details on how to adjust btt’s notion of what constitutes activity. The pairs are arranged as follows: • First there is the total system activity – meaning activity in either queue or completion traces across all devices. 6 http://plasma-gate.weizmann.ac.il/Grace/ “Grace is a WYSIWYG 2D plotting tool for the X Window System and M*tif.” 13 • Next comes per-device activity information – for each device being traced, that request queues Q and C traces are presented. • Last we present pairs per-process. Using this, one is then able to plot regions of activity versus inactivity – and one can gather a sense of deltas between the queueing of IOs and when they are completed. Figure 1 shows a very simplistic chart showing some activity: BTT Sample Q & C Activity Q activity C activity 0 1 2 3 4 5 6 7 8 9 10 Runtime (seconds) Figure 1: Simple Activity Chart When the black line (system Q activity) is high, then the system is seeing relatively continuous incoming queues. Conversely, when it is low, it represents an extended period of time where no queue requests were coming in. Similarly for the red line and C activity. 14 6 Histogram Data Files The histogram data files provide information concerning incoming and outgoing IO sizes (in blocks). For simplicity, the histogram buckets are one-for-one for sizes up to 1,024 blocks in the IO, and then a single bucket for all sizes greater than or equal to 1,024 blocks. The files are again in grace-friendly format, with the first set containing data for the first 1,023 buckets, and a separate set representing sizes ≥ 1024 blocks. (This is done so that one can easily use a separate formatting specification for the latter set.) The first column (X values) is the various IO sizes, and the second column (Y values) represents the number of IOs of that size. Q Histogram Data File Figure 2 is a sample graph generated from data used during some real-world analysis7 . With the visual representation provided by this, one can quickly discern some different characteristics between the 3 runs – in particular, one can see that there is only a single red point (representing 8 blocks per IO), whereas the other two had multiple data points greater than 8 blocks. Q Histogram Chart - AIM7 FServer Number of IOs (log) 4-way IA64 + 4xMSA1000 (w/ 12 disks per MSA) 4194304 2097152 1048576 524288 262144 131072 65536 32768 16384 8192 4096 2048 1024 512 256 128 64 32 16 8 4 2 1 2.6.20 2.6.20-plug 2.6.20-plug-np 0 64 128 192 256 320 384 448 512 576 640 704 Blocks per IO Figure 2: Q Histogram 7 Note the logarithmic nature of the Y axis for this chart. 15 768 832 896 960 4194304 2097152 1048576 524288 262144 131072 65536 32768 16384 8192 4096 2048 1024 512 256 128 64 32 16 8 4 2 1 1024 D Histogram Data File Figure 3 is a sample graph generated from data used during some real-world analysis8 . Again, visually, one can see that the black and blue dots are somewhat similar below about 192 blocks per IO going out. And then one can make the broad generalization of higher reds, lower blues and blacks in the middle. D Histogram Chart - AIM7 FServer Number of IOS (log) 4-way IA64 + 4xMSA1000 (w/ 12 disks per MSA) 2097152 1048576 524288 262144 131072 65536 32768 16384 8192 4096 2048 1024 512 256 128 64 32 16 8 4 2 1 2.6.20 2.6.20-plug 2.6.20-plug-np 0 64 128 192 256 320 384 448 512 576 640 704 Blocks per IO Figure 3: D Histogram 8 Note the logarithmic nature of the Y axis for this chart. 16 768 832 896 960 2097152 1048576 524288 262144 131072 65536 32768 16384 8192 4096 2048 1024 512 256 128 64 32 16 8 4 2 1 1024 7 iostat Data File btt attempts to produce the results from running an iostat -x command in parallel with the system as it is being traced. The fields (columns) generated by the --iostat or -I option can be seen from the following output snippet – note that the line has been split to fit on the printed page: Device: ... ( 8, 16) ... ( 8, 16) rrqm/s rkB/s wrqm/s r/s w/s wkB/s avgrq-sz avgqu-sz 0.00 0.00 0.00 76403.18 1.02 2.02 5.80 27.81 0.00 1005.30 152.00 31.00 0.34 42.13 1.07 0.61 rsec/s await wsec/s svctm %util Stamp 0.00 152806.36 0.00 0.00 0.00 71.79 4.03 0.00 TOTAL 55.62 21.90 0.00 Note that the STAMP field contains the runtime (in seconds) for that line of data. 17 8 Per-IO Data File btt can produce a text file containing time line data for each IO processed. The time line data contains rudimentary information for the following stages: • queue traces • get request traces • insert traces • merge traces • issue traces • completion traces • remap traces The –per-io-dump or -p option triggers this behavior, and will produce a file containing streams of IOs (separated by blank spaces). As an example, here is a snippet of 4 IOs that were merged together, you will note there are 3 merged IOs, and 1 inserted in the stream. The issue and completion traces are replicated per IO. 66,0 : 0.763283556 0.763300157 0.763296365 0.763338848 0.763705760 Q I G D C 6208+8 6208+8 6208+8 6208+32 6208+32 66,0 : 0.763314550 0.763315341 0.763338848 0.763705760 Q M D C 6224+8 6224+8 6208+32 6208+32 66,0 : 0.763321010 0.763321775 0.763338848 0.763705760 Q M D C 6232+8 6232+8 6208+32 6208+32 65,240: 0.763244173 0.763244974 0.763374288 0.763826610 Q M D C 6216+8 6216+8 6208+32 6208+32 The columns provide the following information: 1. Device major/minor. 18 2. Time of the trace (seconds from the start of the run) 3. Trace type 4. start block + number of blocks 19 9 Latency Data Files The latency data files which can be optionally produced by btt provide per-IO latency information, one for queue time (Q2D), one for total IO time (Q2C) and one for latencies induced by lower layer drivers and devices (D2C). In both cases, the first column (X values) represent runtime (seconds), while the second column (Y values) shows the actual latency for a command at that time (either Q2D, D2C or Q2C). 20 10 Seek Data Files btt can also produce two data files containing all IO-to-IO sector deltas, providing seek information which can then be plotted. The produced data file contains 3 sets of data: 1. Combined data – all read and write IOs 2. Read data – just seek deltas for reads 3. Write data – just seek deltas for writes The format of the output file names is to have the name generated by the following fields separated by underscores ( ): • The prefix provided as the argument to the -s option. • The major and minor numbers of the device separated by a comma. • The string q2q or d2d, indicating the Q2Q or D2D seeks, respectively. • One of the following characters: r For read (device to system) IOs w For write (system to device) IOs c Combined – both read and write IOs An example name would be after specifying -s seek would be: seek 065,048 q2q w.dat. The format of the data is to have the runtime values (seconds since the start of the run) in column 1 (X values); and the difference in sectors from the previous IO in column 2 (Y values). Here is a snippet of the first few items from a file: # Combined 0.000034733 0.000106453 0.005239009 0.006968575 0.007218709 0.012145393 0.014980835 0.024239323 0.024249402 0.025707095 ... 35283790.0 35283790.0 35283950.0 35283886.0 35283694.0 35283566.0 -35848914.0 -35848914.0 -35848914.0 -35849072.0 Figure 4 shows a simple graph that can be produced which provides visual details concerning seek patterns. 21 Seek Magnitude BTT Sample Seek Patterns Reads Writes 0 5 10 15 20 Runtime (seconds) Figure 4: Seek Chart The seek difference is calculated in one of two ways: default By default, the seek distance is calculated as the closest distance between the previous IO and this IO. The concept of closeness means that it could either be the end of the previous IO and the beginning of the next, or the end of this IO and the start of the next. -a If the -a or --seek-absolute option is specified, then the seek distance is simply the difference between the end of the previous IO and the start of this IO. 22 10.1 Seeks Per Second When the -m option provides a name, Q2Q and/or D2D seeks will trigger btt to output seeks-per-second information. The first column will contain a time value (seconds), and the second column will indicate the number of seeks per second at that point. When there is only a single data point within a 1-second window, btt will just output the time value for the point, and the value 1.0 in the second column. If there is no perceived difference in the times present for the current sample, then the second columns value is the number of seeks present at that time. Otherwise, if α and Ω are the first and last times seen within a 1-second window, and ν are the number of seeks seen in that time frame, then: column 1 Midway point in time for this span, or: α + (Ω − α)/2 column 2 Average seeks per second over this span, or: ν/(Ω − α) Figure 5 shows a simple pair of graphs generated from -m output: Kernel Make Runs: BTRFS (0.14) 10 passes (untar, make config+all+clean), Linux 2.6.25.4, SATA drive 5000 5000 Seeks per Second Queue 4000 4000 3000 3000 2000 2000 1000 1000 0 0 5000 5000 Issue 4000 4000 3000 3000 2000 2000 1000 1000 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 Runtime (minutes) Figure 5: Seeks-per-second Chart 23 14 15 16 17 0 11 Command Line Usage: btt 2.08 [ -a [ -A [ -B
Source Exif Data:
File Type : PDF File Type Extension : pdf MIME Type : application/pdf PDF Version : 1.2 Linearized : No Page Count : 34 Creator : TeX output 2011.12.10:2049 Producer : dvipdfm 0.13.2d, Copyright © 1998, by Mark A. Wicks Create Date : 2011:12:10 20:49:25-05:00EXIF Metadata provided by EXIF.tools