Date: Sun, 5 Oct 2014 13:19:11 +0200
From: "Steinar H. Gunderson" <sgunderson@bigfoot.com>
To: Chuck Ebbert <cebbert.lkml@gmail.com>
Cc: linux-kernel@vger.kernel.org
Subject: Re: Slowdown due to threads bouncing between HT cores
Message-ID: <20141005111911.GA35329@sesse.net>
References: <20141003194428.GA27084@sesse.net>
 <20141004095004.3b84095f@as>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <20141004095004.3b84095f@as>
User-Agent: Mutt/1.5.21 (2010-09-15)
Sender: linux-kernel-owner@vger.kernel.org

On Sat, Oct 04, 2014 at 09:50:04AM -0500, Chuck Ebbert wrote:
> Try playing with /proc/sys/kernel/sched_migration_cost_ns. This sets
> the number of nanoseconds the kernel will wait before considering
> moving a thread to another CPU. I have mine set to 50000000.

I can't get any good effect out of this. I tried both 50 ms (your value) and
500 ms, and while it seems (by eyeballing the per-cpu display in top) to
_sometimes_ lock the processes to cores, more often than not, they still
bounce around between all 40. Worse still, when it _does_, it seems to often
lock them to hyperthread pairs (e.g., I've seen it put threads only on
virtual cores 0-9 and 20-29, which means it has all of them on one socket!)

Notwithstanding top, the benchmarks don't improve; setting cores manually
with taskset still is much better.

/* Steinar */
-- 
Homepage: http://www.sesse.net/
--
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/