From mboxrd@z Thu Jan 1 00:00:00 1970 Subject: Re: [PATCH 14/14] page-replace-kswapd-incmin.patch From: Peter Zijlstra In-Reply-To: <20051231011507.GC4913@dmt.cnet> References: <20051230223952.765.21096.sendpatchset@twins.localnet> <20051230224212.765.38527.sendpatchset@twins.localnet> <20051231011507.GC4913@dmt.cnet> Content-Type: text/plain Date: Sat, 31 Dec 2005 10:40:40 +0100 Message-Id: <1136022040.17853.11.camel@twins> Mime-Version: 1.0 Content-Transfer-Encoding: 7bit Sender: owner-linux-mm@kvack.org Return-Path: To: Marcelo Tosatti Cc: linux-mm@kvack.org, linux-kernel@vger.kernel.org, Andrew Morton , Christoph Lameter , Wu Fengguang , Nick Piggin , Marijn Meijles , Rik van Riel List-ID: On Fri, 2005-12-30 at 23:15 -0200, Marcelo Tosatti wrote: > On Fri, Dec 30, 2005 at 11:42:34PM +0100, Peter Zijlstra wrote: > > > > From: Nick Piggin > > > > Explicitly teach kswapd about the incremental min logic instead of just scanning > > all zones under the first low zone. This should keep more even pressure applied > > on the zones. > > > > The new shrink_zone() logic exposes the very worst side of the current > > balance_pgdat() function. Without this patch reclaim is limited to ZONE_DMA. > > Can you please describe the issue with over protection of DMA zone you experienced? > > I'll see if I can reproduce it with Nick's standalone patch on top of vanilla, what > load was that? With the mdb bench the following behaviour was observed: (mem=128M) - PageCache would fill zone_normal - PageCache would fill zone_dma - reclaim starts - initially things look right - after a while zone_dma is reclaimed so fast that it frequently gets a full eviction (nr_resident == 0). - from this point onward zone_normal practiaclly sits idle and zone_dma goes wild with all the action. -- Peter Zijlstra -- 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