From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id ; Sun, 8 Apr 2001 10:21:31 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id ; Sun, 8 Apr 2001 10:21:22 -0400 Received: from beton.btnet.cz ([62.80.85.76]:1540 "HELO beton.btnet.cz") by vger.kernel.org with SMTP id ; Sun, 8 Apr 2001 10:21:12 -0400 Date: Sun, 8 Apr 2001 16:19:33 +0200 From: clock@beton.btnet.cz To: linux-kernel@vger.kernel.org Subject: TCP stack misbehaviour? Message-ID: <20010408161933.A223@beton.btnet.cz> Reply-To: clock@ghost.btnet.cz Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline User-Agent: Mutt/1.2.5i Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org The TCP stack in the linux kernel behaves this way: I have got a FULL-DUPLEX 28.800kbps channel with BER <= 1*10-9 a) I start a long TCP connection in one direction b) After 5 minutes I start another connection in the opposite direction The second connection rusn for about half a minute, then converges to 0 throughtput. After several minutes, another 30 seconds of transmission ocuur. The data path in the direction that should be used for this connection is empty, except for occasional ACKs. The utilization of the channel is about 4%. I would expect that both channels would be used for at least 95%. Instead, only one is used. Is this a bug of Linux kernel TCP stack, or a bug in the algorithm presented in the appropriate RFC? Isn't UDP more suitable for data transfers? -- Karel Kulhavy http://atrey.karlin.mff.cuni.cz/~clock