From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id ; Sat, 1 Sep 2001 10:36:50 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id ; Sat, 1 Sep 2001 10:36:41 -0400 Received: from minus.inr.ac.ru ([193.233.7.97]:34573 "HELO ms2.inr.ac.ru") by vger.kernel.org with SMTP id ; Sat, 1 Sep 2001 10:36:30 -0400 From: kuznet@ms2.inr.ac.ru Message-Id: <200109011436.SAA18432@ms2.inr.ac.ru> Subject: Re: Lost TCP retransmission timer To: val@nmt.edu (Val Henson) Date: Sat, 1 Sep 2001 18:36:35 +0400 (MSK DST) Cc: davem@redhat.com, linux-kernel@vger.kernel.org In-Reply-To: <20010830161139.A18224@boardwalk> from "Val Henson" at Aug 30, 1 04:11:39 pm X-Mailer: ELM [version 2.4 PL24] MIME-Version: 1.0 Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Hello! > 2.4.6 machine pushes lots of data on _first_ TCP connection after boot Lots? I see only about 24K of data transmitted in both your samples. Actually, the problem is more or less clear from your /proc/net/tcp. You use some funny device or netfilter plugin, which leak memory. You can look into 7th column of /proc/net/tcp to estimate amount of leaked buffers. When it reaches ~15, connection stalls. Seems, it raises monotonically, so that it looks like all the buffers leak. What is output device? Alexey