From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1759534AbXEQRrU (ORCPT ); Thu, 17 May 2007 13:47:20 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1756979AbXEQRrJ (ORCPT ); Thu, 17 May 2007 13:47:09 -0400 Received: from hancock.steeleye.com ([71.30.118.248]:39457 "EHLO hancock.sc.steeleye.com" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1756373AbXEQRrH (ORCPT ); Thu, 17 May 2007 13:47:07 -0400 Subject: Re: sysfs makes scaling suck Re: Asynchronous scsi scanning From: James Bottomley To: Benjamin LaHaise Cc: Satyam Sharma , Matthew Wilcox , Simon Arlott , Dave Jones , Linux Kernel Mailing List , linux-scsi@vger.kernel.org, kernel-packagers@vger.kernel.org, "Robert P. J. Day" In-Reply-To: <20070517173238.GA30571@kvack.org> References: <1179153096.3703.23.camel@mulgrave.il.steeleye.com> <17841.simon.1179228389@5ec7c279.invalid> <20070515120228.GI10562@parisc-linux.org> <4649E03A.1090004@simon.arlott.org.uk> <20070515172905.GJ10562@parisc-linux.org> <20070517173238.GA30571@kvack.org> Content-Type: text/plain Date: Thu, 17 May 2007 13:45:24 -0400 Message-Id: <1179423924.3785.30.camel@mulgrave.il.steeleye.com> Mime-Version: 1.0 X-Mailer: Evolution 2.8.3 (2.8.3-2.fc6) Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Thu, 2007-05-17 at 13:32 -0400, Benjamin LaHaise wrote: > On Wed, May 16, 2007 at 04:57:52AM +0530, Satyam Sharma wrote: > > > > echo 1 > /sys/module/scsi_mod/.../wait_for_async_scans > > > > somewhere in some script. In fact, the latter method seems simpler, > > saner, better (in every which way)! > > Please don't force sysfs on people. Just watch how it keels over and dies > when you end up with lots of disks/network interfaces on reasonably sized > boxes. 16 statically allocated files per network interface starts being > a problem once you've got a few thousand interfaces. Anything that forces > me to use sysfs is not gonna fly. Well, for that case, you just never enable async scanning But also, the sysfs with over 4,000 (and higher) devices was specifically checked by OSDL (actually as part of the CGL testing) some of the Manoj changes (for unpinning entries etc) were needed to get it to function, but as of now, I believe an enterprise scaling test works reasonably well for it ... there certainly wasn't any evidence of it dying horribly in the tests. James