From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id ; Sat, 11 May 2002 15:55:09 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id ; Sat, 11 May 2002 15:55:08 -0400 Received: from quechua.inka.de ([212.227.14.2]:3610 "EHLO mail.inka.de") by vger.kernel.org with ESMTP id ; Sat, 11 May 2002 15:55:07 -0400 From: Bernd Eckenfels To: linux-kernel@vger.kernel.org Subject: Re: O_DIRECT performance impact on 2.4.18 In-Reply-To: X-Newsgroups: ka.lists.linux.kernel User-Agent: tin/1.5.8-20010221 ("Blue Water") (UNIX) (Linux/2.0.39 (i686)) Message-Id: Date: Sat, 11 May 2002 21:55:06 +0200 Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org In article you wrote: > I bet you could get _better_ performance more cleanly by splitting up the > actual IO generation and the "user-space mapping" thing sanely. For > example, if you want to do an O_DIRECT read into a buffer, there is no > reason why it shouldn't be done in two phases: This works for your load, but it does not work for the load it is designed for. Sequentially reading and througput is not the way to measure it. You need random reading and lattency to see it's merrits. Greetings Bernd