2.6 kernel series have noticeable performance regression, URLs

2.6 kernel series have noticeable performance regression, URLs

[yangyi]

Hi, all

Because of format problem, the previous articles are very hard to read,
now, I can provide a ftp server for downloading those test results.
Please use gedit to read it.

"part 1":
ftp://211.92.88.40/pub/v2.6perf/supermicro-lmbench-lkml
ftp://211.92.88.40/pub/v2.6perf/supermicro-netperf-lkml

"part 2"
ftp://211.92.88.40/pub/v2.6perf/toshiba-lmbench-lkml
ftp://211.92.88.40/pub/v2.6perf/toshiba-netperf-lkml

for other more readable docs, please use this URL:
ftp://211.92.88.40/pub/v2.6perf/

Re: 2.6 kernel series have noticeable performance regression, URLs

[Nick Piggin]

yangyi wrote:
> Hi, all
> 
> Because of format problem, the previous articles are very hard to read,
> now, I can provide a ftp server for downloading those test results.
> Please use gedit to read it.
> 

fork/exec/sh regressions will be mostly due to the unfortunate
4level page tables regressions which have been fixed after 2.6.11.

Also, on the SMP machine, you should bind the entire lmbench test
to just one of the CPUs, because changes to SMP balancing can have
disproportionate or unrealistic changes.

It would be good if you could compare again with the latest 2.6.12
-git release.

Thanks,
Nick

-- 
SUSE Labs, Novell Inc.

Send instant messages to your online friends http://au.messenger.yahoo.com