Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754650AbaLDUwM (ORCPT ); Thu, 4 Dec 2014 15:52:12 -0500 Received: from zeniv.linux.org.uk ([195.92.253.2]:50392 "EHLO ZenIV.linux.org.uk" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753161AbaLDUwK (ORCPT ); Thu, 4 Dec 2014 15:52:10 -0500 Date: Thu, 4 Dec 2014 20:52:02 +0000 From: Al Viro To: Christoph Lameter Cc: Leonard Crestez , Tejun Heo , linux-mm@kvack.org, linux-kernel@vger.kernel.org, Sorin Dumitru Subject: Re: [RFC v2] percpu: Add a separate function to merge free areas Message-ID: <20141204205202.GP29748@ZenIV.linux.org.uk> References: <547E3E57.3040908@ixiacom.com> <20141204175713.GE2995@htj.dyndns.org> <5480BFAA.2020106@ixiacom.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, Dec 04, 2014 at 02:28:10PM -0600, Christoph Lameter wrote: > On Thu, 4 Dec 2014, Leonard Crestez wrote: > > > Yes, we are actually experiencing issues with this. We create lots of virtual > > net_devices and routes, which means lots of percpu counters/pointers. In particular > > we are getting worse performance than in older kernels because the net_device refcnt > > is now a percpu counter. We could turn that back into a single integer but this > > would negate an upstream optimization. > > Well this is not a common use case and that is not what the per cpu > allocator was designed for. There is bound to be signifcant fragmentation > with the current design. The design was for rare allocations when > structures are initialized. ... except that somebody has not known that and took refcounts on e.g. vfsmounts into percpu. With massive amounts of hilarity once docker folks started to test the workloads that created/destroyed those in large amounts. > > Having a "properly scalable" percpu allocator would be quite nice indeed. > > I guess we would be looking at a redesign of the allocator then. FWIW, I think I've already dealt with most of the crap, but I've no idea if networking-related callers end up with similar use patterns. For vfsmounts it seems to suffice... -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/