You can download and get information from http://www.ibm.com/developerworks/wikis/display/WikiPtype/nstress
- I have created 7 different 10g files on all mout points in order to spread IO
dd if=/dev/zero of=/datac1/bigfile1 bs=1m count=10240
dd if=/dev/zero of=/datac2/bigfile2 bs=1m count=10240
dd if=/dev/zero of=/datac3/bigfile3 bs=1m count=10240
dd if=/dev/zero of=/datac4/bigfile4 bs=1m count=10240
dd if=/dev/zero of=/datac5/bigfile5 bs=1m count=10240
dd if=/dev/zero of=/datac6/bigfile6 bs=1m count=10240
dd if=/dev/zero of=/datac7/bigfile7 bs=1m count=10240 - Put the names of the files to a input file
echo "/datac1/bigfile1" >filelist
echo "/datac2/bigfile2">>filelist
echo "/datac3/bigfile3">>filelist
echo "/datac4/bigfile4">>filelist
echo "/datac5/bigfile5">>filelist
echo "/datac6/bigfile6">>filelist
echo "/datac7/bigfile7">>filelist - Identify some system values and also define some assumptions about your system my assumptions and system values are
Block size=8k
Read-WriteRatio: 70:30 = read mostly(OLTP)
Timed duration of the test in seconds =120
Mutliple processes used to generate =3 - Start test for different multiple process
/home/sturgut/ndisk64 -F filelist -S -r70 -b 8k -t 120 -M3
Command: /home/sturgut/ndisk64 -F filelist -S -r70 -b 8k -t 120 -M3
Synchronous Disk test (regular read/write)
No. of processes = 3
I/O type = Sequential
Block size = 8192
Read-WriteRatio: 70:30 = read mostly
Sync type: none = just close the file
Number of files = 7
File size = 33554432 bytes = 32768 KB = 32 MB
Run time = 120 seconds
Snooze % = 0 percent
----> Running test with block Size=8192 (8KB) ...
Proc - <-----disk io----=""> | <-----throughput------> RunTime
Num - TOTAL IO/sec | MB/sec KB/sec Seconds
1 - 2523632 21030.3 | 164.30 168242.34 120.00
2 - 2442193 20351.6 | 159.00 162813.09 120.00
3 - 2097104 17475.9 | 136.53 139807.03 120.00
TOTALS 7062929 58857.8 | 459.83 Seq procs= 3 read= 70% bs= 8KB-----throughput------>-----disk> - In another session Monitor IO service times using # iostat -RDTl
- Increase the different multiple process and monitor the service time .Create a IOPS vs. IO service time chart
- Increase the number of threads to get a peak IOPS
- Be sure your queue_depth is >= number of threads
- More than queue_depth x 2 threads won’t increase thruput
No comments:
Post a Comment