From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S261469AbUBUB0i (ORCPT ); Fri, 20 Feb 2004 20:26:38 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S261470AbUBUB0h (ORCPT ); Fri, 20 Feb 2004 20:26:37 -0500 Received: from fw.osdl.org ([65.172.181.6]:38073 "EHLO mail.osdl.org") by vger.kernel.org with ESMTP id S261469AbUBUB0f (ORCPT ); Fri, 20 Feb 2004 20:26:35 -0500 Date: Fri, 20 Feb 2004 17:28:27 -0800 From: Andrew Morton To: Chad Walstrom Cc: linux-kernel@vger.kernel.org Subject: Re: PROBLEM: NAT over 3c59x cards freezes interactivity on 2.6.2 Message-Id: <20040220172827.09fd7f9a.akpm@osdl.org> In-Reply-To: <20040220224134.GA24203@wookimus.net> References: <20040212171647.GW6864@wookimus.net> <20040212184208.48a93b39.akpm@osdl.org> <20040220224134.GA24203@wookimus.net> X-Mailer: Sylpheed version 0.9.7 (GTK+ 1.2.10; i586-pc-linux-gnu) Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Chad Walstrom wrote: > > On Thu, Feb 12, 2004 at 06:42:08PM -0800, Andrew Morton wrote: > > Chad Walstrom wrote: > > > > > > [2.] DETAILS: Upon upgrading from 2.4.x to 2.6.2, I've noticed that my > > > workstation/NAT box freezes on interactive processes while transfering > > > large files over NAT to other machines on the network. > > > > Please monitor the system time with top or `vmstat 1'. Is it excessive? > > I'm not sure what excessive might mean. There doesn't seem to be any > spikes in memory consumption. Following is the output from 'vmstat 1'. > The number of procs waiting for run time increases when the > interactivity freezes (while the laptop is downloading files over the > workstation's NAT). The '6' and '5' both occur during download attempts > and interactivy freezes. When I kill the download, interactivy resumes, > and the waiting processes drops. > > Script started on Fri Feb 20 16:17:30 2004 > [16:17:30] chewie@skuld (501)$ vmstat 1 > procs -----------memory---------- ---swap-- -----io---- --system-- ----cpu---- > r b swpd free buff cache si so bi bo in cs us sy id wa > 2 0 9364 2560 46256 300188 0 0 10 12 35 44 96 1 3 0 > 1 0 9364 2560 46256 300188 0 0 0 0 1008 63 100 0 0 0 > 1 0 9364 2560 46264 300188 0 0 0 12 1008 56 100 0 0 0 It claims that the CPU is pegged in userspace. What does `top' say? > > A kernel profile might tell us what is going on. See > > Documentation/basic_profiling.txt. > > I'll have to recompile the kernel to test this for you. I'll do so and > compile 2.6.3. Well if you're spending 100% of CPU in userspace then a kernel profile won't tell us much. > > If you remove all the netfilter rules, what happens? > > Then I wouldn't be able to NAT. ;-) The workstation works fine as a > standalone. I've downloaded large files w/o too much difficulty, and > interactivity remains stable. The only time I see problems is when I'm > downloading files through NAT to the laptop. So it is the presence of the NAT rules which is causing the problem? Perhaps the above vmstat trace is wrong? Make sure that you have a nice and recent procps package.