From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from dresden.studentenwerk.mhn.de ([141.84.225.229]:58566 "EHLO email.studentenwerk.mhn.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S2387599AbfBKPla (ORCPT ); Mon, 11 Feb 2019 10:41:30 -0500 From: Wolfgang Walter Subject: Re: [PATCH 1/2] Revert "mm: don't reclaim inodes with many attached pages" Date: Mon, 11 Feb 2019 16:34:34 +0100 Message-ID: <2582452.8YSf1DXvRS@stwm.de> In-Reply-To: <20190207102750.GA4570@quack2.suse.cz> References: <20190130041707.27750-1-david@fromorbit.com> <20190131221904.GL4205@dastard> <20190207102750.GA4570@quack2.suse.cz> MIME-Version: 1.0 Content-Transfer-Encoding: quoted-printable Content-Type: text/plain; charset="iso-8859-1" Sender: linux-xfs-owner@vger.kernel.org List-ID: List-Id: xfs To: Jan Kara Cc: Dave Chinner , Roman Gushchin , Michal Hocko , Chris Mason , linux-mm@kvack.org, linux-kernel@vger.kernel.org, linux-fsdevel@vger.kernel.org, linux-xfs@vger.kernel.org, akpm@linux-foundation.org, vdavydov.dev@gmail.com Am Donnerstag, 7. Februar 2019, 11:27:50 schrieb Jan Kara: > On Fri 01-02-19 09:19:04, Dave Chinner wrote: > > Maybe for memcgs, but that's exactly the oppose of what we want to > > do for global caches (e.g. filesystem metadata caches). We need to > > make sure that a single, heavily pressured cache doesn't evict smal= l > > caches that lower pressure but are equally important for > > performance. > >=20 > > e.g. I've noticed recently a significant increase in RMW cycles in > > XFS inode cache writeback during various benchmarks. It hasn't > > affected performance because the machine has IO and CPU to burn, bu= t > > on slower machines and storage, it will have a major impact. >=20 > Just as a data point, our performance testing infrastructure has bise= cted > down to the commits discussed in this thread as the cause of about 40= % > regression in XFS file delete performance in bonnie++ benchmark. We also bisected our big IO-performance problem of an imap-server (star= ting=20 with 4.19.3) down to =09mm: don't reclaim inodes with many attached pages =09commit a76cf1a474d7dbcd9336b5f5afb0162baa142cf0 upstream. On other servers the filesystems sometimes seems to hang for 10 seconds= and=20 more. We also see a performance regression compared to 4.14 even with this pa= tch=20 reverted, but much less dramatic. Now I saw this thread and I'll try to revert 172b06c32b949759fe6313abec514bc4f15014f4 and see if this helps. Regards, --=20 Wolfgang Walter Studentenwerk M=FCnchen Anstalt des =F6ffentlichen Rechts