Received: with ECARTIS (v1.0.0; list netdev); Wed, 16 Jul 2003 06:03:29 -0700 (PDT) Received: from smtp.istop.com (dci.doncaster.on.ca [66.11.168.194]) by oss.sgi.com (8.12.9/8.12.9) with SMTP id h6GD3EFl018249 for ; Wed, 16 Jul 2003 06:03:14 -0700 Received: from ns.istop.com (ns.istop.com [66.11.168.199]) by smtp.istop.com (Postfix) with ESMTP id 05D9536993; Wed, 16 Jul 2003 08:33:05 -0400 (EDT) Date: Wed, 16 Jul 2003 08:33:04 -0400 (EDT) From: Ralph Doncaster Reply-To: ralph+d@istop.com To: Jeff Garzik Cc: Dan Hollis , linux-netdev Subject: Re: [Bonding-devel] Re: [RFC][bonding] Improve VLAN support on top of bonding In-Reply-To: <3F14807E.30402@pobox.com> Message-ID: References: <3F14807E.30402@pobox.com> MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII X-archive-position: 4103 X-ecartis-version: Ecartis v1.0.0 Sender: netdev-bounce@oss.sgi.com Errors-to: netdev-bounce@oss.sgi.com X-original-sender: ralph@istop.com Precedence: bulk X-list: netdev On Tue, 15 Jul 2003, Jeff Garzik wrote: > Ralph Doncaster wrote: > > On Tue, 15 Jul 2003, Dan Hollis wrote: > > > > > >>That is exactly what it does. hw tcp checksumming helps a LOT at gbe rates > > > > > > This still doesn't make any sense. The copy from user-space to kernel > > space does the checksum as far as I recall (unless you use the > > router-not-host kernel build option). > > > Not for the zero-copy case. How common is this? As far as I can tell, Apache 1.3 doesn't use sendfile (you need 2.0 for that). And even if 1.3 is using EnableMMAP with a large write, you're limited to the size of SO_SNDBUF (or maybe only a single page?). This is not to say hw csum is a bad thing. I think the linux IP stack should support it. When I was looking at the 2.4.19 code I noticed the 3c59x driver code supported hw csum, but I couldn't find anything in the IP stack that used the csum flags set by the driver... -Ralph