Re: [RFC] Low overhead patches for the memory resource controller

[KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>]

On Thu, 14 May 2009 08:09:30 +0530
Balbir Singh <balbir@linux.vnet.ibm.com> wrote:

> * KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> [2009-05-14 09:24:05]:
> 
> > On Thu, 14 May 2009 09:08:02 +0900
> > KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> wrote:
> > 
> > > On Wed, 13 May 2009 21:02:18 +0530
> > > Balbir Singh <balbir@linux.vnet.ibm.com> wrote:
> > > 
> > > > Important: Not for inclusion, for discussion only
> > > > 
> > > > I've been experimenting with a version of the patches below. They add
> > > > a PCGF_ROOT flag for tracking pages belonging to the root cgroup and
> > > > disable LRU manipulation for them
> > > > 
> > > > Caveats:
> > > > 
> > > > 1. I've not checked accounting, accounting might be broken
> > > > 2. I've not made the root cgroup as non limitable, we need to disable
> > > > hard limits once we agree to go with this
> > > > 
> > > > 
> > > > Tests
> > > > 
> > > > Quick tests show an improvement with AIM9
> > > > 
> > > >                 mmotm+patch     mmtom-08-may-2009
> > > > AIM9            1338.57         1338.17
> > > > Dbase           18034.16        16021.58
> > > > New Dbase       18482.24        16518.54
> > > > Shared          9935.98         8882.11
> > > > Compute         16619.81        15226.13
> > > > 
> > > > Comments on the approach much appreciated
> > > > 
> > > > Feature: Remove the overhead associated with the root cgroup
> > > > 
> > > > From: Balbir Singh <balbir@linux.vnet.ibm.com>
> > > > 
> > > > This patch changes the memory cgroup and removes the overhead associated
> > > > with accounting all pages in the root cgroup. As a side-effect, we can
> > > > no longer set a memory hard limit in the root cgroup.
> > > > 
> > > > A new flag is used to track page_cgroup associated with the root cgroup
> > > > pages.
> > > 
> > > Hmm ? How about ignoring memcg completely when the thread belongs to ROOT
> > > cgroup rather than this halfway method ? 
> > > 
> > BTW, this will make softlimit much harder. Do you have any idea on softlimit after
> > this patch ?
> >
> 
> Why would this make soft limit much harder? Since we charge up
> hierarchially even now we ignore a cgroup if its soft limit is not
> set. I am not sure I understand why. 
> 
I doubt I misunderstand something.
Anyway, disabling soft limit to root cgroup is necessary because we have no LRU.

Sorry for noise.

-Kame

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>