Messages in this thread | | | Subject | sysbench(oltp)+mysql 10% regression with 2.6.30-rc1 | From | "Zhang, Yanmin" <> | Date | Fri, 10 Apr 2009 11:01:58 +0800 |
| |
Comparing with 2.6.29, oltp (sysbench+mysql) has 10% regression with 2.6.30-rc1 on 4*2 core stoakley. It's more than 10% on 4*4 core tigerton. Bisect located below patch.
e52fb7c097238d34f4d8e2a596f8a3f85b0c0565 is first bad commit commit e52fb7c097238d34f4d8e2a596f8a3f85b0c0565 Author: Peter Zijlstra <a.p.zijlstra@chello.nl> Date: Wed Jan 14 12:39:19 2009 +0100
sched: prefer wakers
Prefer tasks that wake other tasks to preempt quickly. This improves performance because more work is available sooner.
The workload that prompted this patch was a kernel build over NFS4 (for some curious and not understood reason we had to revert commit: 18de9735300756e3ca9c361ef58409d8561dfe0d to make any progress at all)
Without this patch a make -j8 bzImage (of x86-64 defconfig) would take 3m30-ish, with this patch we're down to 2m50-ish.
psql-sysbench/mysql-sysbench show a slight improvement in peak performance as well, tbench and vmark seemed to not care.
I run the testing with many thread numbers, such like 8, 12 16, 32, 64. The small thread number has no much regression. Most regression is created with thread 32 and thread 64.
If the patch is reverted, the regression becomes very small.
With 2.6.30-rc1, tbench has about 1% improvement on stoakley and 2% on tigerton. If I revert above patch, 1% improvement disappears.
yanmin
| |