About :: TODO :: Blog :: RSS :: Old blog :: Projects :: GIT :: Gallery :: Notes

Tue, 24 Apr 2007

Disk performance emulation.


Benchmark I ran on disk emulator did not include disk seek/read/write time, but now I just use pure loop to perform a delay.
To check how numbers are on the real disk and real filesystem I decided to run yet another filesystem benchmark.
I'm writing set of scripts to automatically run xfs, jfs, ext3/4, reiserfs (maybe reiser4) with set of benchmarks to gather different workload results. I plan to test iozone, postmark and dbench (hopefully I will complete setup and scripts to run it the whole night, but it might not be possible to download archives over my miserable 2kb/s link today).
Hardware is quite small - AMD64 Opteron 3500+ with 1gb of ram, single Maxtor 6E040L0 drive in NFORCE3-250 IDE controller (hdparm shows about 57 MB/s buffered disk read speed).
Stay tuned.

Update1: Postmark from NetApp sucks, I managed to crash it (segfault)... But found a workaround, so will use it.
Update2: Started: postmark for big different sets of file, subdir and transaction numbers, dbench for different number of threads and iozone in automatic mode and with throughput test with different number of threads. All is being logged. All tests are performed for ext4, ext3, ext2, reiserfs, xfs and jfs filesystems. After each test fs is remounted, all caches are dropped. I hope they will be completed this night, so tomorrow I will write bunch of parsing scripts to generate nice graphs and show results here. It is even possible to run such tests in automatic mode to detect regressions in each release/patch.
Ok, now I need to go home and have some sleep...

/devel/fs :: Link / Comments ()