From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from cuda.sgi.com (cuda1.sgi.com [192.48.157.11]) by oss.sgi.com (8.14.3/8.14.3/SuSE Linux 0.8) with ESMTP id qATM6wwZ088862 for ; Thu, 29 Nov 2012 16:06:58 -0600 Received: from ipmail05.adl6.internode.on.net (ipmail05.adl6.internode.on.net [150.101.137.143]) by cuda.sgi.com with ESMTP id usCVbBijJChBMN2z for ; Thu, 29 Nov 2012 14:09:15 -0800 (PST) Date: Fri, 30 Nov 2012 09:09:14 +1100 From: Dave Chinner Subject: Re: [RFC, PATCH 00/19] Numa aware LRU lists and shrinkers Message-ID: <20121129220914.GE6434@dastard> References: <1354058086-27937-1-git-send-email-david@fromorbit.com> MIME-Version: 1.0 Content-Disposition: inline In-Reply-To: List-Id: XFS Filesystem from SGI List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Sender: xfs-bounces@oss.sgi.com Errors-To: xfs-bounces@oss.sgi.com To: Andi Kleen Cc: linux-fsdevel@vger.kernel.org, linux-mm@kvack.org, xfs@oss.sgi.com, linux-kernel@vger.kernel.org, glommer@parallels.com On Thu, Nov 29, 2012 at 11:02:24AM -0800, Andi Kleen wrote: > Dave Chinner writes: > > > > Comments, thoughts and flames all welcome. > > Doing the reclaim per CPU sounds like a big change in the VM balance. It's per node, not per CPU. And AFAICT, it hasn't changed the balance of page cache vs inode/dentry caches under general, global workloads at all. > Doesn't this invalidate some zone reclaim mode settings? No, because zone reclaim is per-node and the shrinkers now can reclaim just from a single node. i.e. the behaviour is now better suited to the aims of zone reclaim which is to free memory from a single, targetted node. Indeed, I removed a hack in the zone reclaim code that sprayed slab reclaim across the entire machine until sufficient objects had been freed from the target node.... > How did you validate all this? fakenuma setups, various workloads that generate even dentry/slab cache loadings across all nodes, adding page cache pressure on a single node, watching slab reclaim from a single node. that sort of thing. I haven't really done any performance testing other than "not obviously slower". There's no point optimising anything before there's any sort of agreement as to whether this is the right approach to take or not.... Cheers, Dave. -- Dave Chinner david@fromorbit.com _______________________________________________ xfs mailing list xfs@oss.sgi.com http://oss.sgi.com/mailman/listinfo/xfs From mboxrd@z Thu Jan 1 00:00:00 1970 From: Dave Chinner Subject: Re: [RFC, PATCH 00/19] Numa aware LRU lists and shrinkers Date: Fri, 30 Nov 2012 09:09:14 +1100 Message-ID: <20121129220914.GE6434@dastard> References: <1354058086-27937-1-git-send-email-david@fromorbit.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: glommer@parallels.com, linux-kernel@vger.kernel.org, linux-fsdevel@vger.kernel.org, linux-mm@kvack.org, xfs@oss.sgi.com To: Andi Kleen Return-path: Content-Disposition: inline In-Reply-To: Sender: owner-linux-mm@kvack.org List-Id: linux-fsdevel.vger.kernel.org On Thu, Nov 29, 2012 at 11:02:24AM -0800, Andi Kleen wrote: > Dave Chinner writes: > > > > Comments, thoughts and flames all welcome. > > Doing the reclaim per CPU sounds like a big change in the VM balance. It's per node, not per CPU. And AFAICT, it hasn't changed the balance of page cache vs inode/dentry caches under general, global workloads at all. > Doesn't this invalidate some zone reclaim mode settings? No, because zone reclaim is per-node and the shrinkers now can reclaim just from a single node. i.e. the behaviour is now better suited to the aims of zone reclaim which is to free memory from a single, targetted node. Indeed, I removed a hack in the zone reclaim code that sprayed slab reclaim across the entire machine until sufficient objects had been freed from the target node.... > How did you validate all this? fakenuma setups, various workloads that generate even dentry/slab cache loadings across all nodes, adding page cache pressure on a single node, watching slab reclaim from a single node. that sort of thing. I haven't really done any performance testing other than "not obviously slower". There's no point optimising anything before there's any sort of agreement as to whether this is the right approach to take or not.... Cheers, Dave. -- Dave Chinner david@fromorbit.com -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@kvack.org. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: email@kvack.org From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1755002Ab2K2WJU (ORCPT ); Thu, 29 Nov 2012 17:09:20 -0500 Received: from ipmail05.adl6.internode.on.net ([150.101.137.143]:49690 "EHLO ipmail05.adl6.internode.on.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753304Ab2K2WJQ (ORCPT ); Thu, 29 Nov 2012 17:09:16 -0500 X-IronPort-Anti-Spam-Filtered: true X-IronPort-Anti-Spam-Result: ApIKAM/bt1B5LN4a/2dsb2JhbABEhUK0R4V7F3OCHgEBBAEOLBwPFAULCAMYLhQlAyETiAoFvyEUjCyDYGEDlgCQRYMGgVEk Date: Fri, 30 Nov 2012 09:09:14 +1100 From: Dave Chinner To: Andi Kleen Cc: glommer@parallels.com, linux-kernel@vger.kernel.org, linux-fsdevel@vger.kernel.org, linux-mm@kvack.org, xfs@oss.sgi.com Subject: Re: [RFC, PATCH 00/19] Numa aware LRU lists and shrinkers Message-ID: <20121129220914.GE6434@dastard> References: <1354058086-27937-1-git-send-email-david@fromorbit.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: User-Agent: Mutt/1.5.21 (2010-09-15) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Thu, Nov 29, 2012 at 11:02:24AM -0800, Andi Kleen wrote: > Dave Chinner writes: > > > > Comments, thoughts and flames all welcome. > > Doing the reclaim per CPU sounds like a big change in the VM balance. It's per node, not per CPU. And AFAICT, it hasn't changed the balance of page cache vs inode/dentry caches under general, global workloads at all. > Doesn't this invalidate some zone reclaim mode settings? No, because zone reclaim is per-node and the shrinkers now can reclaim just from a single node. i.e. the behaviour is now better suited to the aims of zone reclaim which is to free memory from a single, targetted node. Indeed, I removed a hack in the zone reclaim code that sprayed slab reclaim across the entire machine until sufficient objects had been freed from the target node.... > How did you validate all this? fakenuma setups, various workloads that generate even dentry/slab cache loadings across all nodes, adding page cache pressure on a single node, watching slab reclaim from a single node. that sort of thing. I haven't really done any performance testing other than "not obviously slower". There's no point optimising anything before there's any sort of agreement as to whether this is the right approach to take or not.... Cheers, Dave. -- Dave Chinner david@fromorbit.com