From mboxrd@z Thu Jan 1 00:00:00 1970 From: Eric Dumazet Subject: Re: [PATCH net-next-2.6] net: Xmit Packet Steering (XPS) Date: Sat, 21 Nov 2009 07:58:33 +0100 Message-ID: <4B078F99.70100@gmail.com> References: <4B05D8DC.7020907@gmail.com> <87aayg539l.fsf@basil.nowhere.org> <4B07188C.2070808@gmail.com> <20091120223723.GA4460@basil.fritz.box> <65634d660911201642k3930dc78vd576e0e89dc0c794@mail.gmail.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: Andi Kleen , "David S. Miller" , Linux Netdev List To: Tom Herbert Return-path: Received: from gw1.cosmosbay.com ([212.99.114.194]:51734 "EHLO gw1.cosmosbay.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751000AbZKUG6g (ORCPT ); Sat, 21 Nov 2009 01:58:36 -0500 In-Reply-To: <65634d660911201642k3930dc78vd576e0e89dc0c794@mail.gmail.com> Sender: netdev-owner@vger.kernel.org List-ID: Tom Herbert a =E9crit : >=20 > It probably is a special case, but addresses a real problem. Like rp= s, > this addresses the case where a single queue NIC is the bottleneck on= a > system. Anything that takes work off the interrupting core and sprea= ds > it to other cores may alleviate the bottleneck.=20 >=20 Yes, and this brings back an idea I mentioned earlier in April, for a m= ulticast problem we had with AthenaCR. XPS could be 'started' only if caller runs from ksoftirqd, eg we are in= a stress situation, and additional cost of XPS is nothing compared to huge gains= we have to spread part of work to more CPUS. (Costs of XPS includes : ~500 bytes of ICACHE, and IPI)