lkml.org 
[lkml]   [2007]   [Dec]   [21]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: Major regression on hackbench with SLUB (more numbers)
On Fri, 21 Dec 2007, Ingo Molnar wrote:

> > There are patches pending to address these issues. AFAICT Intel is
> > testing if the regression is still there. There is no way for me to
> > verify what is going on there and there is the constant difficulty of
> > getting detailed information about what is going on at Intel. Every
> > couple of month I get a result from that test. Its a really crappy
> > situation where a lot of confusing information is passed around.
>
> of course there is a way to find out, and that's why i mailed you: fix
> the hackbench regression and i'm quite sure you'll improve the TPC-C
> numbers as well. It shows the same kind of overhead in the profile and
> takes just a few seconds to run. Are your pending SLUB patches in
> 2.6.24-rc5-mm1 already?

The tests that I wrote emulate the test behavior that was described to me
by me.

The fixes in 2.6.24-rc5-mm1 improved those numbers. See
http://lkml.org/lkml/2007/10/27/245 which I quoted earlier to you.
However, I have no TPC-C setup here and from what I hear it takes weeks to
run and requires a large support team for tuning.

You can find the slab test suite for that at

http://git.kernel.org/?p=linux/kernel/git/christoph/vm.git;a=shortlog;h=tests

AFAICT the fixes in 2.6.25-rc5-mm1 result in double the alloc performance
(fastpath) of SLAB.

There are fixes that are not merged yet (the cpu alloc patchset) that
seem to make that factor 3 because we can use the segment register to
avoid per cpu array lookups in the fast path.


\
 
 \ /
  Last update: 2007-12-21 22:59    [from the cache]
©2003-2011 Jasper Spaans