From mboxrd@z Thu Jan 1 00:00:00 1970 From: Anton Blanchard Subject: Re: [patch] e1000 TSO parameter Date: Tue, 29 Jul 2003 16:53:07 +1000 Sender: linux-kernel-owner@vger.kernel.org Message-ID: <20030729065307.GC13227@krispykreme> References: <20030714214510.17e02a9f.davem@redhat.com> <16147.37268.946613.965075@napali.hpl.hp.com> <20030714223822.23b78f9b.davem@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: davidm@hpl.hp.com, davidm@napali.hpl.hp.com, scott.feldman@intel.com, linux-kernel@vger.kernel.org, netdev@oss.sgi.com Return-path: To: "David S. Miller" Content-Disposition: inline In-Reply-To: <20030714223822.23b78f9b.davem@redhat.com> List-Id: netdev.vger.kernel.org Hi, > > So we get almost 15% of throughput drop. This was with plain "netkit > > fptd". AFAIK, it does a simple read/write loop (not sendfile()). We've been seeing rather variable results for TSO as well. With TSO off netperf TCP_STREAM will hit line speed and stay there. With TSO on some runs will hit line speed and others will be about 100Mbit/sec slower. > When we use TSO for non-sendfile() applications it really > stresses memory allocations. We do these 64K+ kmalloc()'s > for each packet we construct. Yep we definitely noticed much more higher allocations when watching /proc/slab. Playing around with slab tuning didnt seem to help. Anton