From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752565AbZDNKWu (ORCPT ); Tue, 14 Apr 2009 06:22:50 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1750967AbZDNKWl (ORCPT ); Tue, 14 Apr 2009 06:22:41 -0400 Received: from mail.gmx.net ([213.165.64.20]:39868 "HELO mail.gmx.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with SMTP id S1750746AbZDNKWl (ORCPT ); Tue, 14 Apr 2009 06:22:41 -0400 X-Authenticated: #3373616 X-Provags-ID: V01U2FsdGVkX1+Vkrdo2lOV6n6ZO6CutbSeA3Ogag9/Iuc40TZoVU hL/Gfv4Oyl7t3T Message-ID: <49E463EA.7070407@gmx.de> Date: Tue, 14 Apr 2009 12:22:34 +0200 From: =?ISO-8859-15?Q?Simon_K=E4mpflein?= User-Agent: Thunderbird 2.0.0.21 (X11/20090318) MIME-Version: 1.0 To: linux-kernel@vger.kernel.org Subject: Question: IO priorities for disk writes? Content-Type: text/plain; charset=ISO-8859-15; format=flowed Content-Transfer-Encoding: 7bit X-Y-GMX-Trusted: 0 X-FuHaFi: 0.61 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Hello, is there a possibility to prioritize disk writes? I tried ionice with cfq scheduler, but as far as I understand and tested it only works for reading. Background: I have one block device (SATA drive) and a data stream of about 6 MB/s written by a process to files on the device (fs: reiserfs, fp opened with O_DIRECT). The data is generated externally and processed and buffered in memory by the process. So an average bandwidth of 6 MB/s for writing is required. Another process reads the data in parallel, also with about 6 MB/s. So generally this is no problem when the system has nothing else to do. But when other heavy disk writes occur, e.g. data written to the disk via network (another extreme case: dd if=/dev/zero of=tmpfile), the writing speed drops far below 6 MB/s. So far I tried with no luck (io scheduler cfq): - "ionice -c0 -n0" for the writer and reader - "ionice -c3" for all pdflush threads Is there any possibility to get this working with the current kernel version? The performance of the other processes besides the reader/writer is unimportant. Alternatively I'm thinking about using a second block device for the writing/reading (maybe an USB-Stick?). Best regards, Simon