Re: [PATCH -mmotm 4/5] memcg: avoid oom during recharge at task move

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

On Thu, 3 Dec 2009 15:00:33 +0900
Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> wrote:

> On Thu, 3 Dec 2009 14:22:43 +0900, KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> wrote:
> > On Thu, 3 Dec 2009 13:58:05 +0900
> > Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> wrote:
> > > I'm now trying to decrease these overhead as much as possible, and the current
> > > status is bellow.
> > > 
> > thanks.
> > 
> > > (support for moving swap charge has not been pushed yet in my tree, so I tested
> > > only (1) and (2) cases.)
> > > 
> > >        |  252M  |  512M  |   1G
> > >   -----+--------+--------+--------
> > >    (1) |  0.20  |  0.40  |  0.81
> > >   -----+--------+--------+--------
> > >    (2) |  0.20  |  0.40  |  0.81
> > > 
> > What is the unit of each numbers ? seconds ? And migration of a process with 1G bytes
> > requires 0.8sec ? But, hmm, speed up twice! sounds nice.
> > 
> Ah, these numbers mean "seconds".
> I agree they are big yet...
> 
But maybe reducing this will requires big efforts (or impossible).
So, this number is not very bad I think.


> > 
> > > What I've done are are:
> > > - Instead of calling res_counter_uncharge() against the old cgroup in __mem_cgroup_move_account()
> > >   evrytime, call res_counter_uncharge(PAGE_SIZE * moved) at the end of task migration once.
> > sounds reasonable.
> > 
> > > - Instead of calling try_charge repeatedly, call res_counter_charge(PAGE_SIZE * necessary)
> > >   in can_attach() if possible.
> > sounds reasonable, too.
> > 
> > > - Not only res_counter_charge/uncharge, consolidate css_get()/put() too.
> > > 
> > please do. But, hmm, I'd like to remove css_put/get per pages ;) But I put it aside now.
> > 
> I do agree with you, but removing them would be a big change..
> This change reduced about 0.2sec in 1GB case, so it's a workaround for now.
> 
please go ahead with coalesced css_put()/get().
I agree that workaround is necessary now.

Thanks,
-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>