From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S267437AbUBSXgd (ORCPT ); Thu, 19 Feb 2004 18:36:33 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S267421AbUBSXgc (ORCPT ); Thu, 19 Feb 2004 18:36:32 -0500 Received: from parcelfarce.linux.theplanet.co.uk ([195.92.249.252]:59036 "EHLO www.linux.org.uk") by vger.kernel.org with ESMTP id S267441AbUBSXga (ORCPT ); Thu, 19 Feb 2004 18:36:30 -0500 Message-ID: <40354871.1060009@pobox.com> Date: Thu, 19 Feb 2004 18:36:17 -0500 From: Jeff Garzik User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.4) Gecko/20030703 X-Accept-Language: en-us, en MIME-Version: 1.0 To: MALET JL CC: linux-kernel@vger.kernel.org Subject: Re: [drivers][sata-promise] TX4 has the cache enabled, it should be disabled References: <4031DB3E.8000406@laposte.net> In-Reply-To: <4031DB3E.8000406@laposte.net> Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org MALET JL wrote: > hello, > I've a TX4 150 card and encounter the following effects : > ->"shorts" files (<1Mo) are copied at 35Mo/s av > ->"long" files (>1Mo) are "burst" copied (ie a big "burst" then hangs, > burst, hangs) this has the following effects : > 1) average fall back to 12Mo/s > 2) systems "hangs" (mouse, keyboard behave like a 100% used cpu) but cpu > usage is still slow > 3) mplayer reset the streams every second (cpu usage still low) > > this is 1-2 week that I have this issue, but can't find where it comes > from...... yesterday by chance when I loaded sata-promise I noticed > "write through" this remembered be some issue I go when I first had the > sata board, making a short research made me remember : > > DISABLE THE BOARD CACHE! on windows this is the same when cache is > enabled : performance drops, system interactive is awfull.... There is no on-board cache on the TX4. Further, you are thinking about the SCSI layer's caching support, i.e. basically whether the driver supports SYNCHRONIZE CACHE (aka flush-cache) scsi command. Jeff