mysql-bench tests

[Kent Baxley]

Hello,

I've been trying to find some detailed information on what the mysql-bench tests actually do behind the scenes. Recently, I've had to do some testing of different filesystems with using the mysql-bench test for a customer. They used the mysql-bench tests as proof that xfs, jfs, and reiserfs were all faster than ext3.

The benchmark they ran is the one that comes with MySQL.
They used version 4.1.9 of MySQL RPMs from mysql.com.
All file systems were created with the default settings. No tuning was
done with the FS or the 2.6 kernels. The tests were run using 32bit and 64bit Fedora
Core 3. The production machine will be running in 64bit. On 64bit FC3
the 64bit MySQL RPMs were used.

Nothing else was going on on this machine during the tests.

The following table lists the "TOTAL" output (very last line of the
benchmark output) from the benchmarks:

MySQL-4.1.9 Benchmark DBD suite: 2.15
FS seconds usr sys cpu tests
=======================================
64bit FC3 kernel 2.6.10_1.641_FC3smp + glibc-2.3.4
------------------------------------------------------------------------- ----
XFS 997.00 264.06 99.88 363.94 3425950
JFS 1062.00 266.59 91.81 358.40 3425950
ReiserFS 1086.00 260.46 95.21 355.67 3425950
ext3 1206.00 272.50 91.86 364.36 3425950
------------------------------------------------------------------------- ----
32bit FC3 kernel 2.6.10_1.760_FC3smp + glibc-2.3.4
------------------------------------------------------------------------- ----
XFS 1289.00 347.37 91.96 439.33 3425950
JFS 1320.00 334.10 104.83 438.93 3425950
ReiserFS 1335.00 329.04 102.23 431.27 3425950
ext3 1475.00 345.64 91.75 437.39 3425950

In order to help with tuning ext3 so that it performs as well as (or beats) xfs, I'd like to know more about what these tests are actually doing. I'd expect most database tests to be doing update-in-place IO but I'm not sure. I haven't been able to find any details in the documentation, so any help is appreciated.

Thank you.