From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from cuda.sgi.com (cuda1.sgi.com [192.48.157.11]) by oss.sgi.com (8.14.3/8.14.3/SuSE Linux 0.8) with ESMTP id p3BNXpei117334 for ; Mon, 11 Apr 2011 18:33:55 -0500 Received: from bombadil.infradead.org (localhost [127.0.0.1]) by cuda.sgi.com (Spam Firewall) with ESMTP id 7FA0B14BBC01 for ; Mon, 11 Apr 2011 16:37:11 -0700 (PDT) Received: from bombadil.infradead.org (bombadil.infradead.org [18.85.46.34]) by cuda.sgi.com with ESMTP id VK5GSmyqgGM9OuTQ for ; Mon, 11 Apr 2011 16:37:11 -0700 (PDT) Date: Mon, 11 Apr 2011 19:37:05 -0400 From: Christoph Hellwig Subject: Re: 2 question about XFS fragmentation and _fsr Message-ID: <20110411233705.GA29358@infradead.org> References: <20110411214238.GE21395@dastard> MIME-Version: 1.0 Content-Disposition: inline In-Reply-To: <20110411214238.GE21395@dastard> List-Id: XFS Filesystem from SGI List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Sender: xfs-bounces@oss.sgi.com Errors-To: xfs-bounces@oss.sgi.com To: Dave Chinner Cc: linux-xfs@oss.sgi.com, Janos Haar On Tue, Apr 12, 2011 at 07:42:38AM +1000, Dave Chinner wrote: > Depends on your samba configuration to how it does writes. If it is > conigured to do sync writes then it will have all sorts of > fragmentation problems. So without more details about the workload, > the kernel that is being used, the fragmentation occurring (xfs_bmap > is your friend) and the samba config, there's little that can be > suggested here. Very few people configure Samba to do sync writes, but even without that I see files written by Samba heavily fragmented when the systems operates under intense memory pressure. The CIFS protocol has very small write sizes, and it seems like they do not always arrive in order. If the system is under enough memory pressure to write them out already before the previous fragment arrive it can cause very fragmented extent maps. Enabling the "strict allocate" option in newer Samba version fixes that problem by doing fallocates in larger chunks, which from my observation always happen in order. The downside is that current fallocate in XFS is not designed to be used in a fastpath, and thus causes performance degradations during the actual write. _______________________________________________ xfs mailing list xfs@oss.sgi.com http://oss.sgi.com/mailman/listinfo/xfs