From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 16031 invoked by alias); 10 Sep 2005 17:35:24 -0000 Mailing-List: contact ecos-discuss-help@ecos.sourceware.org; run by ezmlm Precedence: bulk List-Subscribe: List-Archive: List-Post: List-Help: , Sender: ecos-discuss-owner@ecos.sourceware.org Received: (qmail 16013 invoked by uid 22791); 10 Sep 2005 17:35:20 -0000 Received: from sta-204-188-98-138.rockynet.com (HELO hermes.chez-thomas.org) (204.188.98.138) by sourceware.org (qpsmtpd/0.30-dev) with ESMTP; Sat, 10 Sep 2005 17:35:20 +0000 Received: by hermes.chez-thomas.org (Postfix, from userid 2000) id 9C3AD59F17B; Sat, 10 Sep 2005 11:34:58 -0600 (MDT) Received: from localhost.localdomain (localhost.localdomain [127.0.0.1]) by hermes.chez-thomas.org (Postfix) with ESMTP id 171DB59DE98; Sat, 10 Sep 2005 11:34:57 -0600 (MDT) From: Gary Thomas To: Matt Jerdonek Cc: mkhoyila@uci.edu, eCos Discussion In-Reply-To: <20050910172323.41487.qmail@web33514.mail.mud.yahoo.com> References: <20050910172323.41487.qmail@web33514.mail.mud.yahoo.com> Content-Type: text/plain Date: Sat, 10 Sep 2005 17:35:00 -0000 Message-Id: <1126373696.12394.201.camel@hermes> Mime-Version: 1.0 Content-Transfer-Encoding: 7bit Subject: Re: [ECOS] Re: eCos Performance X-SW-Source: 2005-09/txt/msg00055.txt.bz2 On Sat, 2005-09-10 at 10:23 -0700, Matt Jerdonek wrote: > These numbers don't seem reasonable to me. Is there a > chance that something other than the performance of > the ethernet driver is at fault here? For example, > could the RAM be setup with a different number of > wait-states (making the eCos test bed slower). Or > could caching be enabled with Linux but disabled in > eCos? Indeed, this does seem a bit strange. What's the platform/target? How did you configure the network? How did you configure the system? What else is running? Some device drivers have their own tunings (number & size of internal buffers, etc)? If this is so on your driver, how is it configured? One thing to note is that the Linux stack and the eCos stack are tuned for very different environments and are of differing vintages (the Linux design is *years* newer than the BSD design used in eCos). These differences could explain quite a lot. Also, it's normally important to measure performance in something other than a vacuum. Care is taken in the eCos stack not to perturb the real-time characteristics of the system, which may in a micro-sense cause the network to be slower. > > >> Can you help me to improve performance of eCos > >> ethernet driver. I am > >> getting these numbers with similar test setup > > >> Linux: max 26,000 packets (60 bytes) per second > >> (single burst) with no drops > >> eCos: max 8,500 packets (60 bytes) per second > (single > >> burst) with no drops > > >> I was hoping with eCos I could reach around 35,000 > >> packet/sec. Is there > >> any way I could better tune the driver to reach my > >> goal. Thanks. -- ------------------------------------------------------------ Gary Thomas | Consulting for the MLB Associates | Embedded world ------------------------------------------------------------ -- Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss