From mboxrd@z Thu Jan 1 00:00:00 1970 From: Eric Dumazet Subject: RE: [PATCH] mlx4_en: Setting RSS hash result to skb->rxhash field Date: Thu, 05 May 2011 08:57:15 +0200 Message-ID: <1304578635.32152.800.camel@edumazet-laptop> References: <4DC156AF.4050807@mellanox.co.il> <1304520260.3203.34.camel@localhost> <1304527586.32152.5.camel@edumazet-laptop> <953B660C027164448AE903364AC447D2070B8E35@mtldag01.mtl.com> <1304578518.32152.794.camel@edumazet-laptop> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: Ben Hutchings , "davem@davemloft.net" , "netdev@vger.kernel.org" To: Yevgeny Petrilin Return-path: Received: from mail-ww0-f44.google.com ([74.125.82.44]:35093 "EHLO mail-ww0-f44.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751358Ab1EEG5V (ORCPT ); Thu, 5 May 2011 02:57:21 -0400 Received: by wwa36 with SMTP id 36so2017927wwa.1 for ; Wed, 04 May 2011 23:57:19 -0700 (PDT) In-Reply-To: <1304578518.32152.794.camel@edumazet-laptop> Sender: netdev-owner@vger.kernel.org List-ID: Le jeudi 05 mai 2011 =C3=A0 08:55 +0200, Eric Dumazet a =C3=A9crit : > Le jeudi 05 mai 2011 =C3=A0 06:47 +0000, Yevgeny Petrilin a =C3=A9cri= t : > > >=20 > > Eric, you are correct. > > We do plan to enable full 32 bit hash for our devices. > > Once it is done, we will naturally use the whole 32 bits. > > In the meanwhile, even with this change we see improved performance= when enabling RPS. >=20 > Hmm, thats strange because you have only 256 possible values for rxha= sh, > and it's OK for maybe one hundred flows. (I am talking not of RPS but > RFS here) >=20 > So your patch is a win only for small hosts, or particular workloads > (did I say biased benchmarks ? ;) ) >=20 > Really, we better use the linux/software full 32bit rxhash in this ca= se, > and wait for your 32bit full support. >=20 >=20 BTW, before you submit another rxhash patch, please make sure ethtool support is included. An admin must be able to switch off hardware support : ethtool -K eth0 rxhash off