From mboxrd@z Thu Jan 1 00:00:00 1970 From: Tom Herbert Subject: Re: [PATCH v2] Receive Packet Steering Date: Fri, 17 Jul 2009 11:05:58 -0700 Message-ID: <65634d660907171105s4989758ah533203d107924b6a@mail.gmail.com> References: <65634d660907131504u35154059m5934cca3cb9363e0@mail.gmail.com> <20090714.123313.186658126.davem@davemloft.net> <65634d660907141628g671812f9t4219cc8b6a493425@mail.gmail.com> <20090716.194846.58845712.davem@davemloft.net> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: netdev@vger.kernel.org To: David Miller Return-path: Received: from smtp-out.google.com ([216.239.33.17]:4865 "EHLO smtp-out.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1757613AbZGQSGD convert rfc822-to-8bit (ORCPT ); Fri, 17 Jul 2009 14:06:03 -0400 Received: from wpaz29.hot.corp.google.com (wpaz29.hot.corp.google.com [172.24.198.93]) by smtp-out.google.com with ESMTP id n6HI60Gb002556 for ; Fri, 17 Jul 2009 19:06:01 +0100 Received: from qw-out-2122.google.com (qwd5.prod.google.com [10.241.193.197]) by wpaz29.hot.corp.google.com with ESMTP id n6HI5wu3026681 for ; Fri, 17 Jul 2009 11:05:58 -0700 Received: by qw-out-2122.google.com with SMTP id 5so329247qwd.23 for ; Fri, 17 Jul 2009 11:05:58 -0700 (PDT) In-Reply-To: <20090716.194846.58845712.davem@davemloft.net> Sender: netdev-owner@vger.kernel.org List-ID: On Thu, Jul 16, 2009 at 7:48 PM, David Miller wro= te: > > From: Tom Herbert > Date: Tue, 14 Jul 2009 16:28:01 -0700 > > > Using the Toeplitz hash in steering lookup has given us about 10% m= ore > > maximum pps (2 different NICs), and we haven't really noticed negat= ive > > effects because of the extra descriptor overhead-- so I'm not going= to > > give up on it too easily! > > Do you have any idea why? =A0Does Toeplitz distribute better? > If so, that could be merely because either: > > 1) Our modulus avoidance scheme somehow decreases the distribution > =A0 features of the hash > > 2) The way we feed data into the hash has a similar effect > > It's worth checking out. The advantage is that Toeplitz, or any reasonable device provided hash, allows packet steering to be done without taking any cache misses on the packet itself.=A0 In particular, this helps with a NIC that just provides Toeplitz hash (without multiQ),