Message-Id: <20091202091739.5C3D.A69D9226@jp.fujitsu.com>
Date:	Wed,  2 Dec 2009 09:35:19 +0900 (JST)
From:	KOSAKI Motohiro <kosaki.motohiro@...fujitsu.com>
To:	David Rientjes <rientjes@...gle.com>
Cc:	kosaki.motohiro@...fujitsu.com,
	Andrea Arcangeli <aarcange@...hat.com>,
	KAMEZAWA Hiroyuki <kamezawa.hiroyu@...fujitsu.com>,
	linux-mm@...ck.org, linux-kernel@...r.kernel.org,
	Andrew Morton <akpm@...ux-foundation.org>,
	Hugh Dickins <hugh.dickins@...cali.co.uk>,
	vedran.furac@...il.com
Subject: Re: [PATCH] oom_kill: use rss value instead of vm size for badness

> On Tue, 1 Dec 2009, KOSAKI Motohiro wrote:
> 
> > > The purpose of /proc/pid/oom_adj is not always to polarize the heuristic 
> > > for the task it represents, it allows userspace to define when a task is 
> > > rogue.  Working with total_vm as a baseline, it is simple to use the 
> > > interface to tune the heuristic to prefer a certain task over another when 
> > > its memory consumption goes beyond what is expected.  With this interface, 
> > > I can easily define when an application should be oom killed because it is 
> > > using far more memory than expected.  I can also disable oom killing 
> > > completely for it, if necessary.  Unless you have a consistent baseline 
> > > for all tasks, the adjustment wouldn't contextually make any sense.  Using 
> > > rss does not allow users to statically define when a task is rogue and is 
> > > dependent on the current state of memory at the time of oom.
> > > 
> > > I would support removing most of the other heuristics other than the 
> > > baseline and the nodes intersection with mems_allowed to prefer tasks in 
> > > the same cpuset, though, to make it easier to understand and tune.
> > 
> > I feel you talked about oom_adj doesn't fit your use case. probably you need
> > /proc/{pid}/oom_priority new knob. oom adjustment doesn't fit you.
> > you need job severity based oom killing order. severity doesn't depend on any
> > hueristic.
> > server administrator should know job severity on his system.
> 
> That's the complete opposite of what I wrote above, we use oom_adj to 
> define when a user application is considered "rogue," meaning that it is 
> using far more memory than expected and so we want it killed.  As you 
> mentioned weeks ago, the kernel cannot identify a memory leaker; this is 
> the user interface to allow the oom killer to identify a memory-hogging 
> rogue task that will (probably) consume all system memory eventually.  
> The way oom_adj is implemented, with a bit shift on a baseline of 
> total_vm, it can also polarize the badness heuristic to kill an 
> application based on priority by examining /proc/pid/oom_score, but that 
> wasn't my concern in this case.  Using rss as a baseline reduces my 
> ability to tune oom_adj appropriately to identify those rogue tasks 
> because it is highly dynamic depending on the state of the VM at the time 
> of oom.

 - I mean you don't need almost kernel heuristic. but desktop user need it.
 - All job scheduler provide memory limitation feature. but OOM killer isn't
   for to implement memory limitation. we have memory cgroup.
 - if you need memory usage based know, read /proc/{pid}/statm and write
   /proc/{pid}/oom_priority works well probably.
 - Unfortunatelly, We can't continue to use VSZ based heuristics. because
   modern application waste 10x VSZ more than RSS comsumption. in nowadays,
   VSZ isn't good approximation value of RSS. There isn't any good reason to
   continue form desktop user view.

IOW, kernel hueristic should adjust to target majority user. we provide a knob
to help minority user.

or, Can you have any detection idea to distigish typical desktop and your use case?



--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists