From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id ; Thu, 24 Oct 2002 11:58:27 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id ; Thu, 24 Oct 2002 11:58:27 -0400 Received: from dbl.q-ag.de ([80.146.160.66]:37287 "EHLO dbl.q-ag.de") by vger.kernel.org with ESMTP id ; Thu, 24 Oct 2002 11:58:26 -0400 Message-ID: <3DB81A0E.7040000@colorfullife.com> Date: Thu, 24 Oct 2002 18:04:30 +0200 From: Manfred Spraul User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.1) Gecko/20020827 X-Accept-Language: en-us, en MIME-Version: 1.0 To: Roy Sigurd Karlsbakk , "Richard B. Johnson" CC: linux-kernel@vger.kernel.org Subject: "Richard B. Johnson" Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org > > >A 400 MHz ix86 CPU will checksum/copy at 685 megabytes per second. >It will copy at 1,549 megabytes per second. Those are megaBYTES! > > Interesting. You copy data with 1,549 megabytes per second over a 800 megabytes per second frond side bus, into (probably) PC100 memory, i.e. theoretical bandwidth of 800 megabytes/sec. You probably operate within the L2 cache. Roy, which type of memory and which chipset do you use? The csum/copy speed is 210 MByte/sec: 13% cpu time needed for 225 mbit/sec. That could be correct, if you use SDRAM and everything is cache-cold (source dma from disk, dest previously used as a network buffer, dma from nic) -- Manfred