From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S271419AbTHDJHx (ORCPT ); Mon, 4 Aug 2003 05:07:53 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S271420AbTHDJHx (ORCPT ); Mon, 4 Aug 2003 05:07:53 -0400 Received: from main.gmane.org ([80.91.224.249]:30615 "EHLO main.gmane.org") by vger.kernel.org with ESMTP id S271419AbTHDJHw (ORCPT ); Mon, 4 Aug 2003 05:07:52 -0400 X-Injected-Via-Gmane: http://gmane.org/ To: linux-kernel@vger.kernel.org From: mru@users.sourceforge.net (=?iso-8859-1?q?M=E5ns_Rullg=E5rd?=) Subject: Re: 2.6 ide i/o performance Date: Mon, 04 Aug 2003 11:04:42 +0200 Message-ID: References: <200308021205.59280.vt@vt.fermentas.lt> Mime-Version: 1.0 Content-Type: text/plain; charset=iso-8859-1 Content-Transfer-Encoding: 8bit X-Complaints-To: usenet@main.gmane.org User-Agent: Gnus/5.1002 (Gnus v5.10.2) XEmacs/21.4 (Rational FORTRAN, linux) Cancel-Lock: sha1:3Xf1tX1r78ffBbGLQJN3xqQDzSQ= Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Vitalis Tiknius writes: > i simultaneously burned (audio mode, without scsi emulation) and ripped cd's > under 2.6.0-test2-mm2. devices are: > > ../ide/host0/bus1/target0/lun0/cd (Teac CD-W552E) > ../ide/host0/bus1/target1/lun0/cd (Teac DV-516E). > > mobo is Intel 875. software is k3b-0.9, cdrtools-2.01_alpha18, grip-3.1.1, and > cdparanoia-3.9.8 with all paranoia options on. > > when burning and ripping are performed separately, their speeds are approx. > 42x and 6.3x. when simultaneously, 12x and 1.6x with no options touched. > > although devices are on the same controller (my first controller is > SATA) and on the same bus, i'd rather expect linear and not almost > square-law throughput regression observed. are the things expected > to go this way, or there is some room for optimizations, etc.? The first thing to do is to use separate cables for those devices, and see if it helps. -- Måns Rullgård mru@users.sf.net