From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Zhang, Yanmin" Subject: Re: [patch 21/21] slab defrag: Obsolete SLAB Date: Fri, 16 May 2008 13:16:35 +0800 Message-ID: <1210914995.3177.131.camel@ymzhang> References: <20080510030831.796641881@sgi.com> <20080510030919.604216074@sgi.com> <4825709A.2020407@firstfloor.org> <20080510221515.3540a6cc@bree.surriel.com> <2f11576a0805120038s334dc56cuaf16b8b7c6f87098@mail.gmail.com> <84144f020805120054t1370236ei5ff52279457e026e@mail.gmail.com> <482B2617.5010605@firstfloor.org> <1210822002.3177.121.camel@ymzhang> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: Andi Kleen , Pekka Enberg , KOSAKI Motohiro , Rik van Riel , akpm@linux-foundation.org, linux-kernel@vger.kernel.org, linux-fsdevel@vger.kernel.org, Mel Gorman , mpm@selenic.com, Matthew Wilcox To: Christoph Lameter Return-path: Received: from mga06.intel.com ([134.134.136.21]:32567 "EHLO orsmga101.jf.intel.com" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1751239AbYEPFTk (ORCPT ); Fri, 16 May 2008 01:19:40 -0400 In-Reply-To: Sender: linux-fsdevel-owner@vger.kernel.org List-ID: On Thu, 2008-05-15 at 10:05 -0700, Christoph Lameter wrote: > On Thu, 15 May 2008, Zhang, Yanmin wrote: >=20 > > > It can thrash cachelines if objects from the same slab page are f= reed=20 > > > simultaneously on multiple processors. That occurred in the hackb= ench=20 > > > regression that we addressed with the dynamic configuration of sl= ab sizes. > > hackbench regression is because of slow allocation instead of slow = freeing. > > With =C3=BF=C3=BFdynamic configuration of slab sizes, fast allocati= on becomes 97% (the bad > > one is 68%), but fast free is always 8~9% with/without the patch. >=20 > Thanks for using the slab statistics. I wish I had these numbers for = the=20 > TPC benchmark. That would allow us to understand what is going on whi= le it=20 > is running. >=20 > The frees in the hackbench were slow because partial list updates occ= urred=20 > to frequently. The first fix was to let slab sit longer on the partia= l=20 > list.=20 I forgot that. 2.6.24 merged the patch. > The other was the increase of the slab sizes which also increases=20 > the per cpu slab size and therefore the objects allocatable without a= =20 > round trip to the page allocator. That is what I am talking. 2.6.26-rc merged the patch. > Freeing to a per cpu slab never requires=20 > partial list updates. So the frees also benefitted from the larger sl= ab=20 > sizes. But the effect shows up in the count of partial list updates n= ot in=20 > the fast/free collumn. I agree. It might be better if SLUB could be optimized again to =EF=BB=BF= have more consideration when the slow free percentage is high, because the page lock might ping= -pong among processors if multi-processors access the same slab at the same t= ime. -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel= " in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html