From mboxrd@z Thu Jan  1 00:00:00 1970
From: Eric Dumazet <eric.dumazet@gmail.com>
Subject: Re: big picture UDP/IP performance question re 2.6.18 -> 2.6.32
Date: Sat, 01 Oct 2011 21:11:43 +0200
Message-ID: <1317496303.3802.25.camel@edumazet-laptop>
References: <6.2.5.6.2.20111001141002.05af4b20@binnacle.cx>
Mime-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: QUOTED-PRINTABLE
Cc: linux-kernel@vger.kernel.org, netdev <netdev@vger.kernel.org>,
	Peter Zijlstra <a.p.zijlstra@chello.nl>
To: starlight@binnacle.cx
Return-path: <netdev-owner@vger.kernel.org>
Received: from mail-ww0-f44.google.com ([74.125.82.44]:55360 "EHLO
	mail-ww0-f44.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org
	with ESMTP id S1755951Ab1JATLt (ORCPT
	<rfc822;netdev@vger.kernel.org>); Sat, 1 Oct 2011 15:11:49 -0400
In-Reply-To: <6.2.5.6.2.20111001141002.05af4b20@binnacle.cx>
Sender: netdev-owner@vger.kernel.org
List-ID: <netdev.vger.kernel.org>

Le samedi 01 octobre 2011 =C3=A0 14:16 -0400, starlight@binnacle.cx a =C3=
=A9crit :
> At 08:44 AM 10/1/2011 +0200, Eric Dumazet wrote:
> >In my experience, I have the exact opposite :
> >performance greatly improved in recent
> >kernels. Unless you compile your kernel to include
> >new features that might reduce performance
> >(namespaces, cgroup, ...)
>=20
> RH has both of the above turned on in the
> 2.6.32-71.29.1.el6.x86_64 kernel tested.
>=20
> If these are big negatives to network
> performance, could you list what should
> specifically turned off to maximize
> results?  Also a recommendation for
> the best recent kernel for another
> benchmark would be helpful.
>=20
> Probably can't convince anyone to deploy a
> kernel without commercial support, but if
> an alternate compile fixes performance it
> might be possible to convince RH to support
> the alternative build.
>=20

2.6.32 has a perf tool, that can really help to spot in a few minutes
hot spots. That would definitely help to further diagnose what could be
the problem in your workload.

A single patch can have huge performance impact, sometime not noticed.

=46or example, in 2.6.36, AF_UNIX support for pid namespaces dropped
performance a lot [commit 7361c36c5224 (af_unix: Allow credentials to
work across user and pid namespaces)], because of a single atomic
operation, but done on each send() and receive() on a central location.