From mboxrd@z Thu Jan 1 00:00:00 1970 From: Andi Kleen Subject: Re: [rfc][patch] store-free path walking Date: Wed, 7 Oct 2009 23:06:51 +0200 Message-ID: <20091007210651.GB1656@one.firstfloor.org> References: <20091006101414.GM5216@kernel.dk> <20091006122623.GE30316@wotan.suse.de> <20091006124941.GS5216@kernel.dk> <20091007085849.GN30316@wotan.suse.de> <20091007164622.GX30316@wotan.suse.de> <87eipfymcv.fsf@basil.nowhere.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Andi Kleen , Nick Piggin , Jens Axboe , Linux Kernel Mailing List , linux-fsdevel@vger.kernel.org, Ravikiran G Thirumalai , Peter Zijlstra To: Linus Torvalds Return-path: Content-Disposition: inline In-Reply-To: Sender: linux-kernel-owner@vger.kernel.org List-Id: linux-fsdevel.vger.kernel.org On Wed, Oct 07, 2009 at 01:51:07PM -0700, Linus Torvalds wrote: > > > On Wed, 7 Oct 2009, Andi Kleen wrote: > > > > With some strategic prefetches it might be cheap enough to use > > two cache lines. Just fetch it with the other. > > I _seriously_ doubt that. > > This is one of the biggest (and hottest) hash tables in the whole kernel. > > No amount of prefetching will help the fact that you effectively double > your cache footprint / working set if you have to fetch two cachelines in > the hot case lookup rather than one. It won't double it, because there are many more cache lines from all kinds of other things in these paths. The point was just to hide the latency of fetching two at different times. The actual cache foot print is not that important, as long as it's not excessive, just the latencies hurt. -Andi -- ak@linux.intel.com -- Speaking for myself only.