From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S932388AbXDAKVc (ORCPT ); Sun, 1 Apr 2007 06:21:32 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S932418AbXDAKVb (ORCPT ); Sun, 1 Apr 2007 06:21:31 -0400 Received: from noname.neutralserver.com ([70.84.186.210]:42066 "EHLO noname.neutralserver.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932388AbXDAKVa (ORCPT ); Sun, 1 Apr 2007 06:21:30 -0400 Date: Sun, 1 Apr 2007 13:21:28 +0300 From: Dan Aloni To: Pedram M Cc: linux-kernel@vger.kernel.org Subject: Re: Fwd: kswapd issues + kernel 2.4.21-32.0.1.ELsmp Message-ID: <20070401102127.GA16987@localdomain> References: <9c9aa5d00703310110s2a21f19l37544c3d46be53c9@mail.gmail.com> <9c9aa5d00703310111w3809fd56t75bcde9d5047ca81@mail.gmail.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <9c9aa5d00703310111w3809fd56t75bcde9d5047ca81@mail.gmail.com> User-Agent: Mutt/1.5.13 (2006-08-11) X-AntiAbuse: This header was added to track abuse, please include it with any abuse report X-AntiAbuse: Primary Hostname - noname.neutralserver.com X-AntiAbuse: Original Domain - vger.kernel.org X-AntiAbuse: Originator/Caller UID/GID - [0 0] / [47 12] X-AntiAbuse: Sender Address Domain - monatomic.org X-Source: X-Source-Args: X-Source-Dir: Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Sat, Mar 31, 2007 at 01:11:29AM -0700, Pedram M wrote: > Hi, > > I've seen this around, and have heard about it in forums and else-where, > could somebody enlighten me with more information or with experiences > they have had. Looks like kswapd begins to eat CPU dramatically till the > box eventually locks up. My experience with kswapd eating a lot of CPU tells that this particular problem with Linux 2.4 manifests itself on this scenario: * No swap is configured (we are blessed with RAM). _AND_: * A lot of memory is allocated (for my case: a lot of anonymous pages used for applications). Looking at this problem with kdb, I noticed that the CPU spends a lot of time in swap_out(), even though there are no swap devices... So... mm/vmscan.c, replace swap_out with this: static int swap_out(zone_t * classzone) { return 0; } Now, this might raise some eyebrows for those MM gurus reading this mail but I assure you I've been doing this for a lot of time (and a lot of machines) on 2.4.27, and it took care of my kswapd issues quite nicely. I hope it helps... -- Dan Aloni XIV LTD, http://www.xivstorage.com da-x (at) monatomic.org, dan (at) xiv.co.il