Message-ID: <1432210423.2703.54.camel@sauron.fi.intel.com>
Subject: Re: [PATCH RFC] numa,sched: have fbq_classify_* factor out the
 current task
From: Artem Bityutskiy <dedekind1@gmail.com>
Reply-To: dedekind1@gmail.com
To: Rik van Riel <riel@redhat.com>
Cc: linux-kernel@vger.kernel.org, peterz@infradead.org, mgorman@suse.de,
        jhladky@redhat.com, mingo@kernel.org
Date: Thu, 21 May 2015 15:13:43 +0300
In-Reply-To: <20150519105544.68caa726@annuminas.surriel.com>
References: <20150519105544.68caa726@annuminas.surriel.com>
Content-Type: text/plain; charset="UTF-8"
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1225
Lines: 31

On Tue, 2015-05-19 at 10:55 -0400, Rik van Riel wrote:
> The load balancer tries to find, through find_busiest_group and
> find_busiest_queue, a busy CPU with tasks that are running on
> the wrong NUMA node.
> 
> However, the load balancer only moves runnable-but-not-running
> tasks in most situations. This fails horribly when the current
> task on a CPU is on the wrong NUMA node, but the other task(s)
> on the run queue are placed correctly.
> 
> In that situation, what started out as one misplaced tasks
> quickly turns into two misplaced tasks.
> 
> Try to avoid that by factoring out the placement of the current
> task, in order to find groups and runqueues with misplaced tasks
> that are not currently running.
> 
> Signed-off-by: Rik van Riel <riel@redhat.com>

This seem to give small improvement for our eCommerce web workload,
average sever response time went down from ~1.4 to ~1.34. I can run the
workload for longer time to get better numbers.

Thanks,
Artem.

--
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/