From mboxrd@z Thu Jan 1 00:00:00 1970 From: Octavian Purdila Subject: Re: [PATCH][net-next] LRO: improve aggregation in case of zero TSecr packets Date: Mon, 31 Aug 2009 20:15:18 +0300 Message-ID: <200908312015.18470.opurdila@ixiacom.com> References: <200908270208.31581.opurdila@ixiacom.com> <20090830.221105.37473168.davem@davemloft.net> Mime-Version: 1.0 Content-Type: Text/Plain; charset="iso-8859-1" Content-Transfer-Encoding: 7bit Cc: themann@de.ibm.com, raisch@de.ibm.com, eric.dumazet@gmail.com, netdev@vger.kernel.org To: David Miller Return-path: Received: from ixro-out-rtc.ixiacom.com ([92.87.192.98]:7358 "EHLO ixro-ex1.ixiacom.com" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1750929AbZHaRRh (ORCPT ); Mon, 31 Aug 2009 13:17:37 -0400 In-Reply-To: <20090830.221105.37473168.davem@davemloft.net> Sender: netdev-owner@vger.kernel.org List-ID: On Monday 31 August 2009 08:11:05 David Miller wrote: > From: Octavian Purdila > Date: Thu, 27 Aug 2009 02:08:31 +0300 > > > This fixes a temporary performance issue we noticed in back to back > > TSO - LRO tests when such tests are run within five minutes after > > boot. > > > > The TSval field of TCP packets is filled in based on the current > > jiffie, which is initialized at -300*HZ. That means that in 5 minutes > > after reboot it will wrap to zero. > > RFC1323 says we absolutely must ignore zero TSecr values. > > It is a bug that the stack emits a zero value when it means to give a > real TSecr value that will be used. > > Probably we can do something like emit '1' when we would emit '0' > based upon jiffies. > > And this would be an improvement from now in that having a off-by-one > TSecr in this situation is better than emitting one which we can > guarentee will be ignored. Right, why did I thought that the LRO TSecr issue can happen even when emitting a right TSval ? :-/ I'll follow with a patch which takes this approach. Thanks, tavi