From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from sc8-sf-mx2-b.sourceforge.net ([10.3.1.92] helo=mail.sourceforge.net) by sc8-sf-list1-new.sourceforge.net with esmtp (Exim 4.43) id 1Ke8tL-0004mW-Ji for user-mode-linux-devel@lists.sourceforge.net; Fri, 12 Sep 2008 06:40:55 -0700 Received: from jaguar.ecp.fr ([138.195.33.9]) by mail.sourceforge.net with esmtp (Exim 4.44) id 1Ke8tK-00026i-7L for user-mode-linux-devel@lists.sourceforge.net; Fri, 12 Sep 2008 06:40:55 -0700 Received: from localhost (localhost [127.0.0.1]) by jaguar.ecp.fr (Postfix) with ESMTP id 06B326802C for ; Fri, 12 Sep 2008 15:40:45 +0200 (CEST) Received: from jaguar.ecp.fr ([127.0.0.1]) by localhost (jaguar [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id 05290-09 for ; Fri, 12 Sep 2008 15:40:44 +0200 (CEST) Received: from [138.195.30.9] (kheops.system.ecp.fr [138.195.30.9]) by jaguar.ecp.fr (Postfix) with ESMTP id E762C68008 for ; Fri, 12 Sep 2008 15:40:44 +0200 (CEST) Message-ID: <48CA7158.5080500@ecp.fr> Date: Fri, 12 Sep 2008 15:40:40 +0200 From: Nicolas Boullis MIME-Version: 1.0 Subject: [uml-devel] parallel I/O List-Id: The user-mode Linux development list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: quoted-printable Sender: user-mode-linux-devel-bounces@lists.sourceforge.net Errors-To: user-mode-linux-devel-bounces@lists.sourceforge.net To: user-mode-linux-devel@lists.sourceforge.net Hi, I've been using uml on production servers for months, and have been quite happy with it, except for I/O performance. I was using the user-mode-linux package from Debian Etch, based on 2.6.18, and compiled with CONFIG_BLK_DEV_UBD_SYNC. Disabling this option of course helps performance, but I feel that it's somewhat unsafe if the host crashes... Moreover, I don't think it helps read performance. As far as I can see, if one wants both safety and performance, the I/O must not be serialized. So I decided to give it a try. The first step was to run a per-device I/O thread. This improves slightly the performance with several UBD devices: I/O on one device do not block I/O on another device. Moreover, that helps to implement parallelized I/O. Then I managed to run several parallel threads per device. As far as I am concernced, that much improves the performance. But currently, my code is more a dirty proof-of-concept than a clean patch. Would you be interested by my work? Cheers, Nicolas Boullis =C9cole Centrale Paris ------------------------------------------------------------------------- This SF.Net email is sponsored by the Moblin Your Move Developer's challenge Build the coolest Linux based applications with Moblin SDK & win great priz= es Grand prize is a trip for two to an Open Source event anywhere in the world http://moblin-contest.org/redirect.php?banner_id=3D100&url=3D/ _______________________________________________ User-mode-linux-devel mailing list User-mode-linux-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel