From mboxrd@z Thu Jan 1 00:00:00 1970 From: Eric Dumazet Subject: Re: [PATCH 3/4 v2 net-next] net: make GRO aware of skb->head_frag Date: Wed, 02 May 2012 04:45:03 +0200 Message-ID: <1335926703.22133.39.camel@edumazet-glaptop> References: <1335523026.2775.236.camel@edumazet-glaptop> <1335809434.2296.9.camel@edumazet-glaptop> <4F9F21E2.3080407@intel.com> <1335835677.11396.5.camel@edumazet-glaptop> <1335854378.11396.26.camel@edumazet-glaptop> <4FA00C9F.8080409@intel.com> <1335891892.22133.23.camel@edumazet-glaptop> <4FA03D69.6060907@intel.com> Mime-Version: 1.0 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 7bit Cc: Alexander Duyck , David Miller , netdev , Neal Cardwell , Tom Herbert , Jeff Kirsher , Michael Chan , Matt Carlson , Herbert Xu , Ben Hutchings , Ilpo =?ISO-8859-1?Q?J=E4rvinen?= , Maciej =?UTF-8?Q?=C5=BBenczykowski?= To: Alexander Duyck Return-path: Received: from mail-wi0-f178.google.com ([209.85.212.178]:37427 "EHLO mail-wi0-f178.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1758440Ab2EBCpJ (ORCPT ); Tue, 1 May 2012 22:45:09 -0400 Received: by wibhq7 with SMTP id hq7so141322wib.1 for ; Tue, 01 May 2012 19:45:07 -0700 (PDT) In-Reply-To: <4FA03D69.6060907@intel.com> Sender: netdev-owner@vger.kernel.org List-ID: On Tue, 2012-05-01 at 12:45 -0700, Alexander Duyck wrote: > I wasn't worried about the kfree vs put_page, I was worried about the > coalesce case. However, it looks like you are correct and I am not > seeing any issues so everything seems to be working fine. > > I have a hacked together ixgbe up and running now with the new build_skb > logic and RSC/LRO disabled. It looks like it is giving me a 5% > performance boost for small packet routing, but I am using more CPU for > netperf TCP receive tests and I was wondering if you had seen anything > similar on the tg3 driver? > Hmm, tg3 perf for small packets is not changed, because tg3 uses copybreak anyway. #define TG3_RX_COPY_THRESHOLD 256 For big packets, I have no changes in netperf because GRO cannot really provide big enough skbs at 1Gb speed. The changes are noticeable when BDP (Bandwidth*Delay Product) is large and some packet losses happen : You notice lower latencies in TCP stack, because of less collapses (big copies)