From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from relay.sgi.com (relay3.corp.sgi.com [198.149.34.15]) by oss.sgi.com (Postfix) with ESMTP id CC5DB7CA6 for ; Sun, 7 Feb 2016 23:44:34 -0600 (CST) Received: from cuda.sgi.com (cuda1.sgi.com [192.48.157.11]) by relay3.corp.sgi.com (Postfix) with ESMTP id 6E407AC001 for ; Sun, 7 Feb 2016 21:44:34 -0800 (PST) Received: from ipmail06.adl2.internode.on.net (ipmail06.adl2.internode.on.net [150.101.137.129]) by cuda.sgi.com with ESMTP id Fp7GqDLGemTQXuaI for ; Sun, 07 Feb 2016 21:44:23 -0800 (PST) Received: from disappointment.disaster.area ([192.168.1.110] helo=disappointment) by dastard with esmtp (Exim 4.80) (envelope-from ) id 1aSecX-0007Aq-Pn for xfs@oss.sgi.com; Mon, 08 Feb 2016 16:44:21 +1100 Received: from dave by disappointment with local (Exim 4.86) (envelope-from ) id 1aSecX-00021P-P8 for xfs@oss.sgi.com; Mon, 08 Feb 2016 16:44:21 +1100 From: Dave Chinner Subject: [PATCH 0/5 v3] xfs: get rid of xfs_cluster_write() Date: Mon, 8 Feb 2016 16:44:13 +1100 Message-Id: <1454910258-7578-1-git-send-email-david@fromorbit.com> List-Id: XFS Filesystem from SGI List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Errors-To: xfs-bounces@oss.sgi.com Sender: xfs-bounces@oss.sgi.com To: xfs@oss.sgi.com Hi folks, Still clearing out my pending patches. This one was last posted almost 6 months ago: http://oss.sgi.com/archives/xfs/2015-08/msg00523.html The main issue at the time was the fact that the ioends were added to the writepage context and only submitted once write_cached_pages() was done, resulting in building up a large number of writeback pages and allocated ioends before IO submission was done. hence we could potentially prevent writeback progress from being made due to mempool starvation as we weren't following the required rules for the ioend mempool. This is addressed in the new patch 5. instead of chaining ioends onto the writepage context, we submit completed ioends the moment we are done processing the page that triggered allocation of a new ioend. Hence we issue ioends as they are built, rather than as a batch once page processing is done. This ensures that writeback follows the rules required for mempools to work effectively and we won't deadlock due to mempool starvation during low memory writeback. This modified patch set has been sitting in my queue now for almost 4 months, so it's time to get it moved onwards... Cheers, Dave. _______________________________________________ xfs mailing list xfs@oss.sgi.com http://oss.sgi.com/mailman/listinfo/xfs