DomainKey-Signature: a=rsa-sha1; q=dns; c=nofws;
  s=s1024; d=yahoo.com.au;
  h=Received:X-YMail-OSG:From:To:Subject:Date:User-Agent:Cc:References:In-Reply-To:MIME-Version:Content-Type:Content-Transfer-Encoding:Content-Disposition:Message-Id;
  b=N6AaLov+US0tAWUMoncI96+ST7F+mGx5wvpcUI91JuYZBjwYg4+UszdFaxPdAHj51s+VWCGVNFXvgOE51DXZjYdedSna/mMVMhGYxZXzX2uXM9d/CFf2r1PT5Vv0P77nW7INAntQbFd7wY1RLxo1Dly58f5KPLQQvfWP4aL0R7E=  ;
From: Nick Piggin <nickpiggin@yahoo.com.au>
To: Christoph Lameter <clameter@sgi.com>
Subject: Re: SLUB 0:1 SLAB (OOM during massive parallel kernel builds)
Date: Thu, 25 Oct 2007 13:06:38 +1000
User-Agent: KMail/1.9.5
Cc: Alexey Dobriyan <adobriyan@gmail.com>, Mel Gorman <mel@skynet.ie>,
       Pekka Enberg <penberg@cs.helsinki.fi>, linux-kernel@vger.kernel.org,
       linux-mm@vger.kernel.org
References: <20071023181615.GA10377@martell.zuzino.mipt.ru> <200710251234.01440.nickpiggin@yahoo.com.au> <Pine.LNX.4.64.0710241941200.30364@schroedinger.engr.sgi.com>
In-Reply-To: <Pine.LNX.4.64.0710241941200.30364@schroedinger.engr.sgi.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7BIT
Content-Disposition: inline
Message-Id: <200710251306.39237.nickpiggin@yahoo.com.au>
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1583
Lines: 36

On Thursday 25 October 2007 12:43, Christoph Lameter wrote:
> On Thu, 25 Oct 2007, Nick Piggin wrote:
> > > Ummm... all unreclaimable is set! Are you mlocking the pages in memory?
> > > Or what causes this? All pages under writeback? What is the dirty ratio
> > > set to?
> >
> > Why is SLUB behaving differently, though.
>
> Nore sure. Are we really sure that this does not occur using SLAB?

>From the reports it seems pretty consistent. I guess it could well
be something that may occur with SLAB *if the conditions are a bit
different*...

> > Memory efficiency wouldn't be the reason, would it? I mean, SLUB
> > should be more efficient than SLAB, plus have less data lying around
> > in queues.
>
> SLAB may have data around in queues which (if the stars align the right
> way) may allow it to go longer without having to get a page from the page
> allocator.

But page allocs from slab isn't where the OOMs are occurring, so this
seems unlikely (also, the all_unreclaimable logic now should be pretty
strict, so you have to really run the machine out of memory (1GB of
swap gets fully used, then his DMA32 zone is scanned 8 times without
reclaiming a single page).

That said, parallel kernel compiling can really change a lot in memory
footprint depending on small variations in timing. So it might not be
anything to worry about.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/