From mboxrd@z Thu Jan 1 00:00:00 1970 From: Rick Jones Subject: Re: [0/5] GSO: Generic Segmentation Offload Date: Tue, 20 Jun 2006 09:18:45 -0700 Message-ID: <44981FE5.5010004@hp.com> References: <20060620090919.GA31613@gondor.apana.org.au> <20060620093219.GF31854@gondor.apana.org.au> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit Cc: "David S. Miller" , netdev@vger.kernel.org Return-path: Received: from palrel11.hp.com ([156.153.255.246]:26274 "EHLO palrel11.hp.com") by vger.kernel.org with ESMTP id S1751390AbWFTQSs (ORCPT ); Tue, 20 Jun 2006 12:18:48 -0400 To: Herbert Xu In-Reply-To: <20060620093219.GF31854@gondor.apana.org.au> Sender: netdev-owner@vger.kernel.org List-Id: netdev.vger.kernel.org > $ sudo ./ethtool -K lo gso on > $ sudo ifconfig lo mtu 1500 > $ netperf -t TCP_STREAM > TCP STREAM TEST to localhost > Recv Send Send > Socket Socket Message Elapsed > Size Size Size Time Throughput > bytes bytes bytes secs. 10^6bits/sec > > 87380 16384 16384 10.00 3598.17 Would it really mess people up if netperf started doing CPU utilization measurements by default on those platforms where it did not require prior calibrarion? I think that might make it more likely that when folks run tests, even over loopback (esp on MP), we'll get the service demand figures that help show the the change in stack efficiency. rick jones BTW, the style of the netperf test banner tells me you might want to upgrade to a newer version of netperf :)