lkml.org 
[lkml]   [2009]   [Oct]   [12]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 0/2] memcg: improving scalability by reducing lock contention at charge/uncharge
* Ying Han <yinghan@google.com> [2009-10-11 11:34:39]:

> 2009/10/10 KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
>
> > Ying Han wrote:
> > > Hi KAMEZAWA-san: I tested your patch set based on 2.6.32-rc3 but I don't
> > > see
> > > much improvement on the page-faults rate.
> > > Here is the number I got:
> > >
> > > [Before]
> > > Performance counter stats for './runpause.sh 10' (5 runs):
> > >
> > > 226272.271246 task-clock-msecs # 3.768 CPUs ( +-
> > > 0.193%
> > > )
> > > 4424 context-switches # 0.000 M/sec ( +-
> > > 14.418%
> > > )
> > > 25 CPU-migrations # 0.000 M/sec ( +-
> > > 23.077%
> > > )
> > > 80499059 page-faults # 0.356 M/sec ( +-
> > > 2.586%
> > > )
> > > 499246232482 cycles # 2206.396 M/sec ( +-
> > > 0.055%
> > > )
> > > 193036122022 instructions # 0.387 IPC ( +-
> > > 0.281%
> > > )
> > > 76548856038 cache-references # 338.304 M/sec ( +-
> > > 0.832%
> > > )
> > > 480196860 cache-misses # 2.122 M/sec ( +-
> > > 2.741%
> > > )
> > >
> > > 60.051646892 seconds time elapsed ( +- 0.010% )
> > >
> > > [After]
> > > Performance counter stats for './runpause.sh 10' (5 runs):
> > >
> > > 226491.338475 task-clock-msecs # 3.772 CPUs ( +-
> > > 0.176%
> > > )
> > > 3377 context-switches # 0.000 M/sec ( +-
> > > 14.713%
> > > )
> > > 12 CPU-migrations # 0.000 M/sec ( +-
> > > 23.077%
> > > )
> > > 81867014 page-faults # 0.361 M/sec ( +-
> > > 3.201%
> > > )
> > > 499835798750 cycles # 2206.865 M/sec ( +-
> > > 0.036%
> > > )
> > > 196685031865 instructions # 0.393 IPC ( +-
> > > 0.286%
> > > )
> > > 81143829910 cache-references # 358.265 M/sec ( +-
> > > 0.428%
> > > )
> > > 119362559 cache-misses # 0.527 M/sec ( +-
> > > 5.291%
> > > )
> > >
> > > 60.048917062 seconds time elapsed ( +- 0.010% )
> > >
> > > I ran it on an 4 core machine with 16G of RAM. And I modified
> > > the runpause.sh to fork 4 pagefault process instead of 8. I mounted
> > cgroup
> > > with only memory subsystem and start running the test on the root cgroup.
> > >
> > > I believe that we might have different running environment including the
> > > cgroup configuration. Any suggestions?
> > >
> >
> > This patch series is only for "child" cgroup. Sorry, I had to write it
> > clearer. No effects to root.
> >
>
> Ok, Thanks for making it clearer. :) So Do you mind post the cgroup+memcg
> configuration
> while you are running on your host?
>
> Thanks
>

Yes, root was fixed by another patchset now in mainline. Another check
is to see if resource_counter lock shows up in /proc/lock_stats.

--
Balbir


\
 
 \ /
  Last update: 2009-10-12 13:41    [from the cache]
©2003-2011 Jasper Spaans