From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from ipmail05.adl6.internode.on.net ([150.101.137.143]:16902 "EHLO ipmail05.adl6.internode.on.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750917AbcLDVhE (ORCPT ); Sun, 4 Dec 2016 16:37:04 -0500 Date: Mon, 5 Dec 2016 08:36:45 +1100 From: Dave Chinner Subject: Re: [PATCH 0/2] XFS buffer cache scalability improvements Message-ID: <20161204213645.GK11750@dastard> References: <1476821653-2595-1-git-send-email-dev@lynxeye.de> <20161018212116.GC23194@dastard> <20161110230200.GI28922@dastard> <1480715692.6684.1.camel@lynxeye.de> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <1480715692.6684.1.camel@lynxeye.de> Sender: linux-xfs-owner@vger.kernel.org List-ID: List-Id: xfs To: Lucas Stach Cc: linux-xfs@vger.kernel.org On Fri, Dec 02, 2016 at 10:54:52PM +0100, Lucas Stach wrote: ..... > > So with the rhashtable change in place, we've already removed the > > cause of the pag_buf_lock contention (the rbtree pointer chasing) so > > there just isn't any overhead that using RCU can optimise away. > > Hence there's no gains to amortise the efficiency losses using RCU > > freeing introduces, and as a result using RCU is slower than > > traditional locking techniques. > > > > I'll keep testing the rhashtbale code - it look solid enough at this > > point to consider it for the 4.10 cycle. > > > Thanks for running those numbers. I had started to modify the patches > according to the review, but didn't get around to fix the RCU path. > > Do you still consider this change to go in with 4.10? Yes, I posted an up-to-date version of it in this series I posted last friday: https://www.spinics.net/lists/linux-xfs/msg02532.html https://www.spinics.net/lists/linux-xfs/msg02535.html Needs review before I'll merge it for 4.10, though. Cheers, Dave. -- Dave Chinner david@fromorbit.com