From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S933502Ab1FBLCV (ORCPT ); Thu, 2 Jun 2011 07:02:21 -0400 Received: from cantor.suse.de ([195.135.220.2]:59169 "EHLO mx1.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S933181Ab1FBLCU (ORCPT ); Thu, 2 Jun 2011 07:02:20 -0400 Date: Thu, 2 Jun 2011 12:02:16 +0100 From: Mel Gorman To: Bertil Engelholm Cc: "linux-kernel@vger.kernel.org" Subject: Re: PROBLEM: zone_reclaim is hanging high priority real time user pthreads Message-ID: <20110602110216.GF7306@suse.de> References: <980E20D8E719A84BBF593FDE6AEC23DA0AEB3CFAA7@ESESSCMS0351.eemea.ericsson.se> <20110527104823.GA5118@suse.de> <980E20D8E719A84BBF593FDE6AEC23DA0AEB42A245@ESESSCMS0351.eemea.ericsson.se> MIME-Version: 1.0 Content-Type: text/plain; charset=iso-8859-15 Content-Disposition: inline In-Reply-To: <980E20D8E719A84BBF593FDE6AEC23DA0AEB42A245@ESESSCMS0351.eemea.ericsson.se> 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 Fri, May 27, 2011 at 01:22:42PM +0200, Bertil Engelholm wrote: > Thanx for the response. Since a few days back we have tried disabling the > zone reclaim and the system behaves much better so that seems to be the > short term solution we'll go for. Good news. > I also assume that if you have real time pthreads that are sensitive to > stalls you might have to disable zone reclaim also in later kernels > even though the zone reclaim implementation have been radically improved. > It'd be one possibility. However, I understand that at least one person is considering adding an additional level of watermarks that is dependant on the number of real-time threads in the system and their expected usage. The idea would be that latency sensitive applications would be allowed to use a number of pages between two watermarks were other users would wake kswapd or enter direct reclaim. I don't know where that currently stands though. Thanks. -- Mel Gorman SUSE Labs