From mboxrd@z Thu Jan 1 00:00:00 1970 From: Christoph Hellwig Subject: Re: [PATCH 0/4 RESEND] writeback: Dirty list handling changes Date: Fri, 16 May 2014 02:27:20 -0700 Message-ID: <20140516092720.GA27594@infradead.org> References: <1400168517-3565-1-git-send-email-jack@suse.cz> <20140515235514.GH5421@dastard> <20140516004754.GJ5421@dastard> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Jan Kara , linux-fsdevel@vger.kernel.org, Wu Fengguang , Andrew Morton To: Dave Chinner Return-path: Received: from bombadil.infradead.org ([198.137.202.9]:43794 "EHLO bombadil.infradead.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755567AbaEPJ1X (ORCPT ); Fri, 16 May 2014 05:27:23 -0400 Content-Disposition: inline In-Reply-To: <20140516004754.GJ5421@dastard> Sender: linux-fsdevel-owner@vger.kernel.org List-ID: > Seems to me that this problem has been solved elswhere in the > kernel, like for tracking of timers to expire on a given jiffie > (tvec, tvec_base, timer_lists). Perhaps we should be looking to move > to a set of time based lists for efficiently tracking what inodes > should be written back at a given background writeback interval > rather than trying to keep everything on the one list. It's also pretty similar to the mru cache we use for filesystems in XFS, except that we'd start writeback instead of deleting items. The whole bucket scheme there should also help with I/O batching.