[lkml]   [2008]   [Nov]   [16]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [PATCH -mm] vmscan: bail out of page reclaim after swap_cluster_max pages
KAMEZAWA Hiroyuki wrote:
> On Sun, 16 Nov 2008 16:38:56 +0900 (JST)
> KOSAKI Motohiro <> wrote:
>> One more point.
>>> Sometimes the VM spends the first few priority rounds rotating back
>>> referenced pages and submitting IO. Once we get to a lower priority,
>>> sometimes the VM ends up freeing way too many pages.
>>> The fix is relatively simple: in shrink_zone() we can check how many
>>> pages we have already freed and break out of the loop.
>>> However, in order to do this we do need to know how many pages we already
>>> freed, so move nr_reclaimed into scan_control.
>> IIRC, Balbir-san explained the implemetation of the memcgroup
>> force cache dropping feature need non bail out at the past reclaim
>> throttring discussion.

Yes, for we used that for force_empty() in the past, but see below

>> I am not sure about this still right or not (iirc, memcgroup implemetation
>> was largely changed).
>> Balbir-san, Could you comment to this patch?
> I'm not Balbir-san but there is no "force-cache-dropping" feature now.
> (I have no plan to do that.)
> But, mem+swap controller will need to modify reclaim path to do "cache drop
> first" becasue the amount of "mem+swap" will not change when "mem+swap" hit
> limit. It's now set "sc.may_swap" to 0.

Yes, there have been several changes to force_empty() and its meaning, including
movement of accounts. Since you've made most of the recent changes, your
comments are very relevant.

> Hmm, I hope memcg is a silver bullet to this kind of special? workload in
> long term.

:-) From my perspective, hierarchy, soft limits (sharing memory when there is no
contention), some form of over commit support and getting swappiness to work
correctly are very important for memcg.


 \ /
  Last update: 2008-11-17 04:49    [W:0.031 / U:7.324 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site