Testing Windows IO with SQLIO and SysBench


To benchmark IO on Linux and MySQL transaction processing, SysBench is a popular choice that can do both. After poking around at the source code, it seems PostgreSQL and Oracle are also included for transaction processing testing if you have the proper header files, but I didn’t test those.

To benchmark IO on Windows and SQL Server transaction processing, Microsoft provides two tools, SQLIO and SQLIOSim. SQLIO is a misnomer in that it really doesn’t have much to do with SQL Server. It is a general purpose disk IO benchmark tool.

So today I was playing with SysBench and noticed that I can compile and build it on Windows as well. I decided I should run IO benchmark on a single machine with both tools (SQLIO and SysBench), and see if I could reconcile the results.

To make things simple, I thought I would just benchmark random read of 3G (orders of magnitude bigger than disk controller cache) files for 5 minutes (300 seconds) with a single thread using 16Kb block size, without adding any additional queue. I tested this on both my laptop and an Amazon EC2 instance. The commands for both tools are listed below, and they should perform the same thing, as far as I can tell. Let me know if you have any comments/pointers or if I missed anything.

SysBench commands:
[sourcecode language=”text”]
sysbench –test=fileio –file-total-size=3G prepare
sysbench.exe –test=fileio –file-total-size=3G –file-test-mode=rndrd –max-time=300 run
[/sourcecode]
Fro SQLIO, here is the line in param.txt and command used:
[sourcecode language=”text”]
c:\testfile.dat 1 0x0 3072
sqlio -kR -s300 -dc -b16 -frandom -Fparam.txt
[/sourcecode]

As this is a quick test, I ran the same test twice and took the average value for comparison purposes. The detailed output is pasted at the end of this post.

On my Windows XP Pro Service Pack 2 laptop with Intel X-25 SSD:

IO/Second Throughput/Second
SQLIO 3833.5 59.90Mb
SysBench 3390.77 52.98Mb

So on my laptop, SQLIO’s results are 13% higher than that of SysBench.

On Amazon EC2 ami-c3e40daa with EBS device running Windows Server 2008 Datacenter Edition Service Pack 2, whose results varied widely between my two runs:

IO/Second Throughput/Second
SQLIO 678.91 10.61Mb
SysBench 408.96 6.39Mb

On this machine, SQLIO results are 66% higher than that of SysBench.

Below is the gory details.

Here are the detailed output on my laptop:
SQLIO
C:\Program Files\SQLIO>sqlio -kR -s300 -dc -b16 -frandom -Fparam.txt
sqlio v1.5.SG
parameter file used: param.txt
file c:\testfile.dat with 1 thread (0) using mask 0x0 (0)
1 thread reading for 300 secs from file c:\testfile.dat
using 16KB random IOs
using specified size: 3072 MB for file: c:\testfile.dat
initialization done
CUMULATIVE DATA:
throughput metrics:
IOs/sec: 3835.39
MBs/sec: 59.92

C:\Program Files\SQLIO>sqlio -kR -s300 -dc -b16 -frandom -Fparam.txt
sqlio v1.5.SG
parameter file used: param.txt
file c:\testfile.dat with 1 thread (0) using mask 0x0 (0)
1 thread reading for 300 secs from file c:\testfile.dat
using 16KB random IOs
using specified size: 3072 MB for file: c:\testfile.dat
initialization done
CUMULATIVE DATA:
throughput metrics:
IOs/sec: 3832.00
MBs/sec: 59.87

SysBench
C:\MessAround\sysbench-0.4.12\sysbench-0.4.12\sysbench\RelWithDebInfo>sysbench.e
xe –test=fileio –file-total-size=3G –file-test-mode=rndrd –max-time=300 run
sysbench 0.4: multi-threaded system evaluation benchmark

Running the test with following options:
Number of threads: 1

Extra file open flags: 0
128 files, 24Mb each
3Gb total file size
Block size 16Kb
Number of random requests for random IO: 10000
Read/Write ratio for combined random IO test: 1.50
Periodic FSYNC enabled, calling fsync() each 100 requests.
Calling fsync() at the end of test, Enabled.
Using synchronous I/O mode
Doing random read test
Threads started!
WARNING: Operation time (18446744073709226000.000000) is greater than maximal co
unted value, counting as 10000000000000.000000
WARNING: Percentile statistics will be inaccurate
Done.

Operations performed: 10000 Read, 0 Write, 0 Other = 10000 Total
Read 156.25Mb Written 0b Total transferred 156.25Mb (52.143Mb/sec)
3337.16 Requests/sec executed

Test execution summary:
total time: 2.9966s
total number of events: 10000
total time taken by event execution: 2.9343
per-request statistics:
min: 0.01ms
avg: 0.29ms
max: 18446744073709.47ms
approx. 95 percentile: 0.48ms

Threads fairness:
events (avg/stddev): 10000.0000/0.00
execution time (avg/stddev): 2.9343/0.00

C:\MessAround\sysbench-0.4.12\sysbench-0.4.12\sysbench\RelWithDebInfo>sysbench.e
xe –test=fileio –file-total-size=3G –file-test-mode=rndrd –max-time=300 run
sysbench 0.4: multi-threaded system evaluation benchmark

Running the test with following options:
Number of threads: 1

Extra file open flags: 0
128 files, 24Mb each
3Gb total file size
Block size 16Kb
Number of random requests for random IO: 10000
Read/Write ratio for combined random IO test: 1.50
Periodic FSYNC enabled, calling fsync() each 100 requests.
Calling fsync() at the end of test, Enabled.
Using synchronous I/O mode
Doing random read test
Threads started!
WARNING: Operation time (18446744073694841000.000000) is greater than maximal co
unted value, counting as 10000000000000.000000
WARNING: Percentile statistics will be inaccurate
Done.

Operations performed: 10000 Read, 0 Write, 0 Other = 10000 Total
Read 156.25Mb Written 0b Total transferred 156.25Mb (53.818Mb/sec)
3444.38 Requests/sec executed

Test execution summary:
total time: 2.9033s
total number of events: 10000
total time taken by event execution: 2.8777
per-request statistics:
min: 0.01ms
avg: 0.29ms
max: 18446744073696.34ms
approx. 95 percentile: 15.39ms

Threads fairness:
events (avg/stddev): 10000.0000/0.00
execution time (avg/stddev): 2.8777/0.00

Here are the detailed output from Amazon EC2 ami-c3e40daa with EBS device:
SQLIO
c:\Program Files\SQLIO>sqlio -kR -t1 -s300 -dC -frandom -b16 -Fparam.txt -BH -LS

sqlio v1.5.SG
using system counter for latency timings, 3579545 counts per second
parameter file used: param.txt
file c:\testfile.dat with 1 thread (0) using mask 0x0 (0)
1 thread reading for 300 secs from file c:\testfile.dat
using 16KB random IOs
buffering set to use hardware disk cache (but not file cache)
size of file c:\testfile.dat needs to be: 3221225472 bytes
current file size: 0 bytes
need to expand by: 3221225472 bytes
expanding c:\testfile.dat … done.
using specified size: 3072 MB for file: c:\testfile.dat
initialization done
CUMULATIVE DATA:
throughput metrics:
IOs/sec: 1230.94
MBs/sec: 19.23
latency metrics:
Min_Latency(ms): 0
Avg_Latency(ms): 0
Max_Latency(ms): 204
histogram:
ms: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24+
%: 98 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

c:\Program Files\SQLIO>sqlio -kR -t1 -s300 -dC -frandom -b16 -Fparam.txt -BH -LS

sqlio v1.5.SG
using system counter for latency timings, 3579545 counts per second
parameter file used: param.txt
file c:\testfile.dat with 1 thread (0) using mask 0x0 (0)
1 thread reading for 300 secs from file c:\testfile.dat
using 16KB random IOs
buffering set to use hardware disk cache (but not file cache)
using specified size: 3072 MB for file: c:\testfile.dat
initialization done
CUMULATIVE DATA:
throughput metrics:
IOs/sec: 126.88
MBs/sec: 1.98
latency metrics:
Min_Latency(ms): 0
Avg_Latency(ms): 7
Max_Latency(ms): 497
histogram:
ms: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24+
%: 13 9 0 3 7 8 8 8 8 8 8 8 2 1 1 1 1 1 1 1 1 0 0 0 2

C:\Users\Administrator\Documents\sysbench-0.4.12\sysbench\RelWithDebInfo>sysbenc
h.exe –test=fileio –file-total-size=3G –file-test-mode=rndrd –max-time=300 r
un
sysbench 0.4: multi-threaded system evaluation benchmark

Running the test with following options:
Number of threads: 1

Extra file open flags: 0
128 files, 24Mb each
3Gb total file size
Block size 16Kb
Number of random requests for random IO: 10000
Read/Write ratio for combined random IO test: 1.50
Periodic FSYNC enabled, calling fsync() each 100 requests.
Calling fsync() at the end of test, Enabled.
Using synchronous I/O mode
Doing random read test
Threads started!
Done.

Operations performed: 10000 Read, 0 Write, 0 Other = 10000 Total
Read 156.25Mb Written 0b Total transferred 156.25Mb (10.64Mb/sec)
680.95 Requests/sec executed

Test execution summary:
total time: 14.6854s
total number of events: 10000
total time taken by event execution: 14.6048
per-request statistics:
min: 0.01ms
avg: 1.46ms
max: 150.29ms
approx. 95 percentile: 4.77ms

Threads fairness:
events (avg/stddev): 10000.0000/0.00
execution time (avg/stddev): 14.6048/0.00

C:\Users\Administrator\Documents\sysbench-0.4.12\sysbench\RelWithDebInfo>sysbenc
h.exe –test=fileio –file-total-size=3G –file-test-mode=rndrd –max-time=300 r
un
sysbench 0.4: multi-threaded system evaluation benchmark

Running the test with following options:
Number of threads: 1

Extra file open flags: 0
128 files, 24Mb each
3Gb total file size
Block size 16Kb
Number of random requests for random IO: 10000
Read/Write ratio for combined random IO test: 1.50
Periodic FSYNC enabled, calling fsync() each 100 requests.
Calling fsync() at the end of test, Enabled.
Using synchronous I/O mode
Doing random read test
Threads started!
Done.

Operations performed: 10000 Read, 0 Write, 0 Other = 10000 Total
Read 156.25Mb Written 0b Total transferred 156.25Mb (2.1371Mb/sec)
136.77 Requests/sec executed

Test execution summary:
total time: 73.1139s
total number of events: 10000
total time taken by event execution: 73.0284
per-request statistics:
min: 0.02ms
avg: 7.30ms
max: 728.84ms
approx. 95 percentile: 23.08ms

Threads fairness:
events (avg/stddev): 10000.0000/0.00
execution time (avg/stddev): 73.0284/0.00

, , , ,

5 responses to “Testing Windows IO with SQLIO and SysBench”

  1. Looks like you’re comparing apples with oranges:)
    e.g SQLIO is using one file and sysbench is using 128 files for the test, there might be other differences, like filesystem caching(sysbench uses it by default), async io (sysbench does not use it at all).

  2. Thanks Mark and Vladislav.

    >>SQLIO is using one file and sysbench is using 128 files for the test
    Good point, Vladislav. I did another test with SysBench with one single file 3G in size.

    sysbench –test=fileio –file-total-size=3G –file-num=1 prepare
    sysbench 0.4: multi-threaded system evaluation benchmark

    1 files, 3145728Kb each, 3072Mb total
    Creating files for the test…

    And then run the benchmark twice on the laptop.
    sysbench –test=fileio –file-total-size=3G –file-test-mode=rndrd –max-time=300 –file-num=1 run

    This time the SysBench shows more IOPS and bigger throughput,
    1st run: Total transferred 156.25Mb (88.97Mb/sec) 5694.09 Requests/sec executed
    2nd run: Total transferred 156.25Mb (88.34Mb/sec) 5653.73 Requests/sec executed

    I would argue this is more like comparing oranges with tangerines, if you know what I mean 🙂 The experiment serves 2 purposes: 1 it satisfies my curiosities; 2 it serves as a safety/sanity check to know that the results are close enough, within the realm of possibilities; that I am not doing things that are totally stupid but for whatever reason, I fail to see it myself.

    For hardware evaluation and capacity planning, it would makes more sense to compare results measured from the same tool set.

  3. hi,
    could you please explain how to use sysbench on windows.
    I tried the steps on http://www.spidernetromania.net/forum/thread-19353-post-316593.html#pid316593, but i get an error.

    Problem is that in the last step, i get the error: “‘sysbench’ is not recognized as an internal or external command”
    When is enter “sysbench.sln” at he command prompt, Visual studio opens. I don’t know how to use sysbench in Visrual studio.
    Does anyone one what the problem is, and how i can run sysbench?

    Other question:
    is there a an .exe available for Windows server 2008 r2? Will that solve this problem?

    Thanks in advance!

    regards,
    Charles

  4. Hi team,

    Can you please send sysbench.exe files to me.

    Thanks for your valuable response.

    Thanks,
    Dhayalan B

Leave a Reply

Your email address will not be published.

This site uses Akismet to reduce spam. Learn how your comment data is processed.