lkml.org 
[lkml]   [2004]   [Oct]   [25]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [patch] scheduler: active_load_balance fixes
From
Date
On Sun, 2004-10-24 at 02:37 -0700, Andrew Morton wrote:
> Nick Piggin <piggin@cyberone.com.au> wrote:
> >
> >
> >
> > Darren Hart wrote:
> >
> > >The following patch against the latest mm fixes several problems with
> > >active_load_balance().
> > >
> > >
> >
> > This seems much better. Andrew can you put this into -mm please.
> >
>
> Whenever we touch the load balancing we get sad little reports about
> performance regressions two months later. How do we gain confidence in
> this change?
>

I did run a kernbench test and forgot to include the results, my
apologies. On a 16 way NUMA the new code was slightly faster. The new
code basically does what I believe the original code was intended to do,
it doesn't take a radically new approach to load balancing. It closes
up some corner cases (like continuing to balance after the runqueue is
empty) and fragile code (like removing the dependency on the order of
the sched_group construction). It also removes some of the artificial
limits imposed by the old code, like always moving tasks to the last CPU
of a completely idle group.

I will run some more tests on this code today to improve our confidence.
It would help if others could do the same, particularly those who have
experience balancing problems in the past.

Thanks,

--
Darren Hart
IBM, Linux Technology Center
503 578 3185
dvhltc@us.ibm.com
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 14:07    [from the cache]
©2003-2011 Jasper Spaans