From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id ; Mon, 7 May 2001 13:02:45 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id ; Mon, 7 May 2001 13:02:35 -0400 Received: from m327-mp1-cvx1c.col.ntl.com ([213.104.77.71]:4992 "EHLO [213.104.77.71]") by vger.kernel.org with ESMTP id ; Mon, 7 May 2001 13:02:18 -0400 To: Manfred Spraul Cc: Subject: Re: [PATCH] zero^H^H^H^Hsingle copy pipe In-Reply-To: <3AF3FA84.AA76CF89@colorfullife.com> <3AF45FAB.595951F8@colorfullife.com> <3AF473F8.F01EF82A@colorfullife.com> From: John Fremlin Date: 07 May 2001 18:01:35 +0100 In-Reply-To: <3AF473F8.F01EF82A@colorfullife.com> Message-ID: User-Agent: Gnus/5.0808 (Gnus v5.8.8) XEmacs/21.1 (GTK) MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org [Stuff about NetBSD pipes snipped] I'm testing out Manfred's patch for zero copy pipes, and haven't crashed it yet. My hardware is a AMD K6-2 (stepping 1) on an ALi M1541 with 320 Mb - one quite slow 64 Mb stick and one fast 256 Mb stick. The lmbench bw_pipe showed a performance improvement of about 30% from 45 (+- 2) Mb/s to 59.5 (+-0.6) Mb/s. The lmbench (2beta3) lat_pipe showed a performance improvement of about 20% from a latency of about 27 (+- 1) usec to about 22.4 (+-.6) usec. There was one outlyer amoung the 10 non zc pipe runs - 25 usec. For zc, the first run was always about 25 usec and after that very stable around 22 usec. FWIW the system time from "time" when running (bw_pipe;lat_pipe) 10 times in a row *increased* by 50%, from sys 0m18.740s to sys 0m31.660s. Script: for i in 1 2 3 4 5 6 7 8 9 10; do ./bw_pipe; ./lat_pipe; done Non zero copy: real 0m49.602s user 0m10.170s sys 0m18.740s Zero copy run 1: real 0m47.901s user 0m10.390s sys 0m31.660s Zero copy run 2: real 0m47.492s user 0m10.600s sys 0m31.340s -- http://ape.n3.net