Date: Wed, 2 Aug 2017 11:22:55 +0100
From: Mel Gorman <mgorman@suse.de>
To: riel@redhat.com
Cc: linux-kernel@vger.kernel.org, peterz@infradead.org, mingo@kernel.org,
        jhladky@redhat.com, lvenanci@redhat.com
Subject: Re: [RHEL-ALT-7.4 PATCH 2/2] sched,numa: scale scan period with
 tasks in group and shared/private
Message-ID: <20170802102255.4tc7jpyt52kjj7tq@suse.de>
References: <20170731192847.23050-1-riel@redhat.com>
 <20170731192847.23050-3-riel@redhat.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-8859-15
Content-Disposition: inline
In-Reply-To: <20170731192847.23050-3-riel@redhat.com>
User-Agent: NeoMutt/20170421 (1.8.2)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1150
Lines: 31

On Mon, Jul 31, 2017 at 03:28:47PM -0400, Rik van Riel wrote:
> From: Rik van Riel <riel@redhat.com>
> 
> Running 80 tasks in the same group, or as threads of the same process,
> results in the memory getting scanned 80x as fast as it would be if a
> single task was using the memory.
> 
> This really hurts some workloads.
> 

It would be nice to specify what workloads in particular and what sort
of machine because I'm willing to bet it has a bigger impact on machines
with 4+ nodes, particularly if they are not fully connected topologies.
Furthermore, I'm willing to bet that there would be small regressions on
2-socket machines but with less time spent scanning and processing
faults even if remote accesses are marginally increased.

Still, on balance, this is preferred behaviour.

> Scale the scan period by the number of tasks in the numa group, and
> the shared / private ratio, so the average rate at which memory in
> the group is scanned corresponds roughly to the rate at which a single
> task would scan its memory.
> 
> Signed-off-by: Rik van Riel <riel@redhat.com>

Acked-by: Mel Gorman <mgorman@suse.de>

-- 
Mel Gorman
SUSE Labs