Re: [RFD] Isolated memory cgroups again

On Thursday, October 20, 2011, KAMEZAWA Hiroyuki <
kamezawa.hiroyu@jp.fujitsu.com> wrote:

On Thu, 20 Oct 2011 19:45:55 -0700
Michal Hocko [off-list ref] wrote:

quoted

On Thu 20-10-11 16:41:27, Ying Han wrote:
[...]

quoted

Hi Michal:

Hi,

quoted

I didn't read through the patch itself but only the description. If we
wanna protect a memcg being reclaimed from under global memory
pressure, I think we can approach it by making change on soft_limit
reclaim.

I have a soft_limit change built on top of Johannes's patchset, which
does basically soft_limit aware reclaim under global memory pressure.

Is there any link to the patch(es)? I would be interested to look at
it before we discuss it.

I'd like to see it, too.

Thanks,
-Kame

Now I am at airport heading to Prague , I will try to post one before the
meeting if possible. The current patch is simple enough which most of the
work are reverting the existing soft limit implementation and then the new
logic is based on the memcg aware global reclaim.

The logic is based on reclaim priority, and we skip reclaim from certain
memcg(under soft limit) before getting down to DEF_PRIORITY - 3. This is
simple enough to get us start collecting some data result and I am looking
forward to discuss more thoughts in the meeting


--ying