[RFC 1/4] memcg: drop pages at rmdir (v1)

KAMEZAWA Hiroyuki kamezawa.hiroyu at jp.fujitsu.com
Tue May 27 17:12:37 PDT 2008


On Tue, 27 May 2008 21:41:26 +0530
Balbir Singh <balbir at linux.vnet.ibm.com> wrote:

> KAMEZAWA Hiroyuki wrote:
> > Now, when we remove memcg, we call force_empty().
> > This call drops all page_cgroup accounting in this mem_cgroup but doesn't
> > drop pages. So, some page caches can be remaind as "not accounted" memory
> > while they are alive. (because it's accounted only when add_to_page_cache())
> > If they are not used by other memcg, global LRU will drop them.
> > 
> > This patch tries to drop pages at removing memcg. Other memcg will
> > reload and re-account page caches. (but this will increase page-in
> > after rmdir().)
> > 
> 
> The approach seems fair, but I am not sure about the overhead of flushing out
> cached pages. Might well be worth it.
> 
> > Consideration: should we recharge all pages to the parent at last ?
> >                But it's not precise logic.
> > 
> 
> We should look into this - I should send out the multi-hierarchy patches soon.
Yes. I'll write my version (if I can). please pick it up if you like it.

> We should discuss this after that.
>
ok. I'd like to forget this patch for a while.

Thanks,
-Kame 








More information about the Containers mailing list