From mboxrd@z Thu Jan 1 00:00:00 1970 From: Octavian Purdila Subject: Re: network card with LRO (and TSO) in hardware? Date: Thu, 29 May 2008 21:11:01 +0300 Message-ID: <200805292111.01690.opurdila@ixiacom.com> References: <200805292004.58064.opurdila@ixiacom.com> <483EE9B3.80201@hp.com> Mime-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: 7bit Cc: netdev@vger.kernel.org To: Rick Jones Return-path: Received: from ixia01.ro.gtsce.net ([212.146.94.66]:2999 "EHLO ixro-ex1.ixiacom.com" rhost-flags-OK-FAIL-OK-FAIL) by vger.kernel.org with ESMTP id S1753153AbYE2SMN (ORCPT ); Thu, 29 May 2008 14:12:13 -0400 In-Reply-To: <483EE9B3.80201@hp.com> Content-Disposition: inline Sender: netdev-owner@vger.kernel.org List-ID: On Thursday 29 May 2008, Rick Jones wrote: > > That second bit sounds very much like a "how long is a piece of string" > question. What size packets? What sort of host system(s)? How many > concurrent flows/connections etc etc etc... > No restrictions... Any packet size (though I presume the lower the better), any kind of host systems, any number of concurrent flows (though I presume the fewer the better). I am just wondering what is the maximum packet rate ever seen by somebody, with hardware LRO. I am asking this because, theoretically, with hardware LRO we should be able to achieve a very high packet rate when using a low MTU (like 64 or 128 bytes). With software LRO I got a significant PPS increase for TCP traffic - from 100K (that with terrible hacks - generating traffic from kernel) to 140K (that cleanly, from user-space). So I am wondering what the hardware LRO will do. Thanks, tavi