Re: Performance regressions in 2.6.30-rc7?

From: Jan Kara
Date: Wed Jun 03 2009 - 15:58:31 EST


> On Sábado 30 Mayo 2009 06:07:18 Andrew Morton escribió:
> > Those regressions are huge. They appear to be on compute-intensive
> > workloads too.
> >
> > I wonder if some expensive debugging option turned itself on or
> > something silly like that.
>
> They use the vanilla kernels that ubuntu ship with their own configs,
> so they shouldn't be using any expensive debugging option.
>
> I've run myself two of those tests (I've not tested gnupg as I'm
> allergic to it). I can't reproduce any regression with 7zip, I'm however
> able to reproduce the regression in tiobench write latency, using 32
> threads and 64 MB block size (--threads 32 --block 65536). I used ext4
> (phoronix uses ext3).
>
>
> Random Writes
> File Blk Num Avg Maximum Lat% Lat% CPU
> Identifier Size Size Thr Rate (CPU%) Latency Latency >2s >10s Eff
> ---------------------------- ------ ----- --- ------ ------ --------- ----------- -------- -------- -----
> 2.6.29 2000 65536 32 24.16 107.2% 0.051 0.21 0.00000 0.00000 23
>
> Random Writes
> File Blk Num Avg Maximum Lat% Lat% CPU
> Identifier Size Size Thr Rate (CPU%) Latency Latency >2s >10s Eff
> ---------------------------- ------ ----- --- ------ ------ --------- ----------- -------- -------- -----
> 2.6.30-rc7-00149-g3218911 2000 65536 32 23.06 178.4% 0.711 164.72 0.00000 0.00000 13
Hmm, also the CPU utilization seems to be much higher with comparable
(in fact slightly lower) throughput. Would it be possible to get
oprofile from the run on 2.6.30 and 2.6.29?

Honza
--
Jan Kara <jack@xxxxxxx>
SuSE CR Labs
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/