From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S262991AbTI2VJ0 (ORCPT ); Mon, 29 Sep 2003 17:09:26 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S261957AbTI2VJ0 (ORCPT ); Mon, 29 Sep 2003 17:09:26 -0400 Received: from law12-f95.law12.hotmail.com ([64.4.19.95]:22788 "EHLO hotmail.com") by vger.kernel.org with ESMTP id S262991AbTI2VJY (ORCPT ); Mon, 29 Sep 2003 17:09:24 -0400 X-Originating-IP: [213.117.178.114] X-Originating-Email: [a1b2c3d4_66@hotmail.com] From: =?iso-8859-1?B?bWFyaW8gY2FtdfFhcw==?= To: linux-kernel@vger.kernel.org Date: Mon, 29 Sep 2003 21:09:23 +0000 Mime-Version: 1.0 Content-Type: text/plain; charset=iso-8859-1; format=flowed Message-ID: X-OriginalArrivalTime: 29 Sep 2003 21:09:23.0443 (UTC) FILETIME=[F5109C30:01C386CD] Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Hello: Our scenario is the following: One web server runninng kernel 2.4.9-e-12(smp), it is used to host an Apache webserver with an integrated application server. The system has 2Gb of RAM and 2Gb of swap. The last day an strange thing happened. We tried to connect to our system using ssh and we canīt, after many attempts we could connect to the system and saw that we couldnīt connect because there was a lack of memory. Every proccess you tried to start failed and gave the following error message" fork failed: Canīt allocate memory (errno=12)". The output of the free -k command showed that the swap space(there was only 5 Mb of free RAM) wasnīt being used and we donīt understand why. We have found a lot of similar cases of systems that couldnīt fork but were plenty of unused swap space so we thought we would find answers for this problem but we havenīt found any(although we have looked for one a lot). We think that perhaps our freepages settings are too low( 1.6-4.5-7.4 Mb) and if we merge this low values with an excesive fragmented memory this could explain our memory squeeze problems. Some of us have proposed the theory that everytime a proccess is started it needs a little quantity of contiguous memory and that if the system canīt provide it to the process it dies before the fork is completed. Any idea about what can be happening here? Regards, Mario. _________________________________________________________________ Descubre el mayor catálogo de coches de la Red en MSN Motor. http://motor.msn.es/researchcentre/