From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jan Kara Subject: Re: [PATCH 0/7 v4] ext4: Extent status tree shrinker improvements Date: Wed, 3 Dec 2014 09:21:18 +0100 Message-ID: <20141203082118.GA8526@quack.suse.cz> References: <1416927336-9328-1-git-send-email-jack@suse.cz> <20141203022018.GM13618@thunk.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Jan Kara , linux-ext4@vger.kernel.org, Zheng Liu To: Theodore Ts'o Return-path: Received: from cantor2.suse.de ([195.135.220.15]:51052 "EHLO mx2.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751925AbaLCIVT (ORCPT ); Wed, 3 Dec 2014 03:21:19 -0500 Content-Disposition: inline In-Reply-To: <20141203022018.GM13618@thunk.org> Sender: linux-ext4-owner@vger.kernel.org List-ID: On Tue 02-12-14 21:20:18, Ted Tso wrote: > I realized I hadn't ack'ed this patch series, but since this patch > series is an improvement on what we have now, I've included them in > the ext4 git tree. I'm sure there's more improvements we can do on > the shrinker, but we can use these patches as a starting point. Yeah, I'd like to return to this as well. In particular it should be relatively straightforward to RCUify the list of inodes with extents to shrink and thus avoid the relatively expensive s_es_lock in the shrinker. Additionally we could batch removal from the list of inodes with reclaimable extents (removal would happen only in the ext4_evict_inode() and in the shrinker when it sees say 16 inodes with zero reclaimable extents in the list - then it would grab the lock, scan part of the list and remove all inodes with zero reclaimable extents). I'd expect that to significantly reduce contention on s_es_lock. But these need more measurements and experimentation... Honza -- Jan Kara SUSE Labs, CR