Date: Mon, 7 Sep 2009 12:42:31 +0200
From: Jens Axboe <jens.axboe@oracle.com>
To: Richard Kennedy <richard@rsk.demon.co.uk>
Cc: Andrew Morton <akpm@linux-foundation.org>,
       Wu Fengguang <fengguang.wu@intel.com>,
       "chris.mason" <chris.mason@oracle.com>, linux-mm <linux-mm@kvack.org>,
       lkml <linux-kernel@vger.kernel.org>,
       Peter Zijlstra <a.p.zijlstra@chello.nl>,
       linux-fsdevel <linux-fsdevel@vger.kernel.org>
Subject: Re: [RFC PATCH] v2 mm: balance_dirty_pages.  reduce calls to
	global_page_state to reduce cache references
Message-ID: <20090907104231.GR18599@kernel.dk>
References: <1252062330.2271.61.camel@castor> <20090906184214.GL18599@kernel.dk> <1252318290.2348.20.camel@castor>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <1252318290.2348.20.camel@castor>
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1724
Lines: 44

On Mon, Sep 07 2009, Richard Kennedy wrote:
> On Sun, 2009-09-06 at 20:42 +0200, Jens Axboe wrote:
> > On Fri, Sep 04 2009, Richard Kennedy wrote:
> > > Reducing the number of times balance_dirty_pages calls global_page_state
> > > reduces the cache references and so improves write performance on a
> > > variety of workloads.
> > > 
> > > 'perf stats' of simple fio write tests shows the reduction in cache
> > > access.
> > > Where the test is fio 'write,mmap,600Mb,pre_read' on AMD AthlonX2 with
> > > 3Gb memory (dirty_threshold approx 600 Mb)
> > > running each test 10 times, dropping the fasted & slowest values then
> > > taking 
> > > the average & standard deviation
> > > 
> > > 		average (s.d.) in millions (10^6)
> > > 2.6.31-rc8	648.6 (14.6)
> > > +patch		620.1 (16.5)
> > 
> > This patch looks good to me, I have workloads too here where up to 10%
> > of the time is spent in balance_dirty_pages() because of this. I'll give
> > this patch a go on the box and test in question tomorrow, but it looks
> > promising.
> > 
> 
> Thanks Jens, 
> 
> It will be interesting to see how it works on different hardware &
> workload. How many cores are you going to run it on?
> wow 10% in balance_dirty_pages! Is that on a large server? or do you
> think its peculiar to your workload?

~10% is rounded, it was 8.9x%. Pretty bad! It's a 64-way, and there's
nothing special about the workload (just buffered IO with 4k blocks).
It's just going rather fast.

-- 
Jens Axboe

--
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/