From mboxrd@z Thu Jan 1 00:00:00 1970 From: Ben Hutchings Subject: Re: Ethernet CRC questions Date: Wed, 30 Dec 2009 17:29:35 +0100 Message-ID: <1262190575.5685.8.camel@localhost> References: <1262108376.5941.5.camel@localhost> <1262169433.2247.15.camel@localhost> Mime-Version: 1.0 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 7bit Cc: netdev@vger.kernel.org To: thomas yang Return-path: Received: from mail.solarflare.com ([216.237.3.220]:40592 "EHLO exchange.solarflare.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752394AbZL3Q3l (ORCPT ); Wed, 30 Dec 2009 11:29:41 -0500 In-Reply-To: Sender: netdev-owner@vger.kernel.org List-ID: On Thu, 2009-12-31 at 00:19 +0800, thomas yang wrote: > >> > >> Does the Ethernet frame CRC (generate and verify in hardware) make > >> the TX / RX rate a little smaller , and a longer end-to-end delay ? > > > > Relative to what? > > > > Relative to the Ethernet frame without CRC appending / verifying . > > > In IEEE 802.3 standard, all network devices should append/verify CRC? Not 'should' but 'must'. > >> Could I disable CRC appending / verifying in the network card driver > >> on all of my nodes (machines) ? > > > > Maybe if you want switches to discard everything you send... > > > > So, I should disable CRC appending / verifying on all kinds of network > devices in my network? > In this way I will TX/RX packets more faster ? You will probably not be able to disable CRC validation on switches because it is not a sensible thing to do. So if you disable CRC generation on NICs then your network will notwork. Really, 32 bit-times is not much to worry about (only 320 ns even at 100M). There are likely to be much bigger sources of latency to attack. Ben. -- Ben Hutchings, Senior Software Engineer, Solarflare Communications Not speaking for my employer; that's the marketing department's job. They asked us to note that Solarflare product names are trademarked.