From mboxrd@z Thu Jan 1 00:00:00 1970 From: Eric Dumazet Subject: Re: TCP_STREAM performance regression on commit b3613118 Date: Tue, 06 Mar 2012 04:01:28 -0800 Message-ID: <1331035288.2474.33.camel@edumazet-laptop> References: <1329472239.2861.3.camel@edumazet-HP-Compaq-6005-Pro-SFF-PC> <20120217.133327.1178765872497293871.davem@davemloft.net> <1330656317.21053.1411.camel@debian> <20120301.220743.2174015405223036036.davem@davemloft.net> <20120306081117.GA17375@feng-i7> <4F55CA25.2070503@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: Feng Tang , David Miller , alex.shi@intel.com, linux-kernel@vger.kernel.org, tim.c.chen@intel.com, ying.huang@intel.com, netdev@vger.kernel.org To: Jason Wang Return-path: In-Reply-To: <4F55CA25.2070503@redhat.com> Sender: linux-kernel-owner@vger.kernel.org List-Id: netdev.vger.kernel.org Le mardi 06 mars 2012 =C3=A0 16:26 +0800, Jason Wang a =C3=A9crit : > On 03/06/2012 04:11 PM, Feng Tang wrote: > > On Thu, Mar 01, 2012 at 10:07:43PM -0500, David Miller wrote: > >> > From: Alex Shi > >> > Date: Fri, 02 Mar 2012 10:45:17 +0800 > >> > =20 > >>> > > Add CC to tang feng, He is working on this issue. > >> > =20 > >> > Is he? I'm pretty sure this is due to the TCP receive window g= rowing > >> > issue Eric Dumazet, Neal Cardwell and I are discussing in the t= hread > >> > starting at: > >> > =20 > >> > http://marc.info/?l=3Dlinux-netdev&m=3D132916352815286&w=3D2 > > Yes, probably, as we did find some clue related with the tcp_r/wmem= =2E > > > > Here is the regression we found: > > On some machines, we found there is about 10% resgression of netper= f > > TCP-64K loopback test between 3.2 and 3.3-rc1. The exact test is: > > ./netperf -t TCP_STREAM -l 60 -H 127.0.0.1 -i 50,3 -I 99,5 -- -s 32= 768 -S 32768 -m 4096 > > > > > > The test machine is a 2 socket Quad Core Core 2 Duo server(2.66GHz)= with > > 8 GB RAM. Following are the debug info (ifconfig/netstat -s/tcp_rwm= em) > > before and after the test: > > > > The most obvious differences I can see are: > > 1) 311 GB vs 241 GB from ifconfig > > 2) the difference of the tcp_r/wmem >=20 > Hi: >=20 > Could you try the newest kernel? Looks like the difference has been=20 > already fixed by commit c43b874d5d714f271b80d4c3f49e05d0cbf51ed2. >=20 Most likely yes.=20 tcp_rmem 4096 87380 87380 tcp_wmem 4096 16384 65536 Is way pessimistic :(