Description:
While testing of the latest MySQL server releases (5.0.27 and 5.1.12) with dbt2 benchmark on the SuSE 10.1(x86_64) performance degradation was observed. Results obtained on 10.1 looked 'unpredictable'. Running the test for the same binaries on the same hardware but on SuSE 9.2 Enterprise we've got much better results. Below are example of results of 4 runs for each OS for MySQL 5.0.27 with 80 db threads. Results are transaction per second, bigger is better:
SuSE 9.2: 18967.18, 19001.29, 18984.04, 18972.12
SuSE 10.1: 11010.71, 9920.32, 11732.66, 10177.28
Preliminary analysis of system information gathered during testing(iostat, vmstat, sar) showed that for SuSE 10.1 we have a higher IO and lower CPU utilization than for 9.2.
HW details: 2xDual Core Opteron 270(AMD64), 6Gb RAM, SCSI disks.
OSes details:
fs in both OSes: ext3
9.2:
# getconf GNU_LIBC_VERSION
glibc 2.3.3
# getconf GNU_LIBPTHREAD_VERSION
NPTL 0.61
# uname -r
2.6.5-7.191-smp
10.1:
# getconf GNU_LIBC_VERSION
glibc 2.4
# getconf GNU_LIBPTHREAD_VERSION
NPTL 2.4
# uname -r
2.6.16.13-4-smp
How to repeat:
run the dbt2 test on SuSE 9.2 and 10.1 on similar HW