To: David Rientjes <rientjes@google.com>
Cc: David Miller <davem@davemloft.net>, eric.dumazet@gmail.com,
       jpiszcz@lucidpixels.com, linux-kernel@vger.kernel.org
References: <alpine.DEB.2.00.0906180935130.2894@chino.kir.corp.google.com>
	<20090618.120034.246557840.davem@davemloft.net>
	<alpine.DEB.2.00.0906181212380.10979@chino.kir.corp.google.com>
	<20090618.123708.78849607.davem@davemloft.net>
	<m1ljnoyn0e.fsf@fess.ebiederm.org>
	<alpine.DEB.2.00.0906191524470.30455@chino.kir.corp.google.com>
From: ebiederm@xmission.com (Eric W. Biederman)
Date: Fri, 19 Jun 2009 18:28:49 -0700
In-Reply-To: <alpine.DEB.2.00.0906191524470.30455@chino.kir.corp.google.com> (David Rientjes's message of "Fri\, 19 Jun 2009 15\:37\:17 -0700 \(PDT\)")
Message-ID: <m1hbybwv5a.fsf@fess.ebiederm.org>
User-Agent: Gnus/5.11 (Gnus v5.11) Emacs/22.2 (gnu/linux)
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Subject: Re: [patch] ipv4: don't warn about skb ack allocation failures
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1584
Lines: 40

David Rientjes <rientjes@google.com> writes:

> On Fri, 19 Jun 2009, Eric W. Biederman wrote:
>
>> But Mr. Reintjes if you really want the traces I can set up 
>> an script to email them to you every time it happens.  Say about
>> 1 a minute from my paltry little farm of machines.
>> 
>
> Perhaps you missed my email where I suggested emitting the page allocation 
> warnings only when CONFIG_DEBUG_VM is enabled.  It's at 
> http://lkml.org/lkml/2009/6/18/355.
>
> We can then keep the __GFP_NOFAIL flag to indicate that the warnings 
> should never be emitted for that allocation, regardless of the .config.
>
> It's funny, though, that the problem that originally started this thread 
> was quickly diagnosed because of these messages.  As far as I know, my 
> suggestion to increase /proc/sys/vm/dirty_background_ratio to kick pdflush 
> earlier has prevented the slab allocation failures and not required 
> delayed acks for nfsd.

increase?

Perhaps then the problem is simply dirty_background_ratio.  Is the vm
not properly autotuning?

With a 50MB/s disk I wonder what the proper window size is.  Several
gigabytes as implied by a 5% or a 10% dirty_background_ratio seems
absurd.  TCP sockets seem to get along fine with even large latencies
and windows measured in megabytes not gigabytes.

This does explain a few things.

Eric
--
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/