From mboxrd@z Thu Jan 1 00:00:00 1970 From: Rick Jones Subject: Re: [PATCH] make _minimum_ TCP retransmission timeout configurable take 2 Date: Thu, 30 Aug 2007 18:07:13 -0700 Message-ID: <46D769C1.8090808@hp.com> References: <200708310009.RAA04175@tardy.cup.hp.com> <20070830.173912.79067694.davem@davemloft.net> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit Cc: netdev@vger.kernel.org To: David Miller Return-path: Received: from palrel12.hp.com ([156.153.255.237]:39639 "EHLO palrel12.hp.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753656AbXHaBHb (ORCPT ); Thu, 30 Aug 2007 21:07:31 -0400 In-Reply-To: <20070830.173912.79067694.davem@davemloft.net> Sender: netdev-owner@vger.kernel.org List-Id: netdev.vger.kernel.org David Miller wrote: > From: Rick Jones > Date: Thu, 30 Aug 2007 17:09:04 -0700 (PDT) > > >>Enable configuration of the minimum TCP Retransmission Timeout via >>a new sysctl "tcp_rto_min" to help those who's networks (eg cellular) >>have quite variable RTTs avoid spurrious RTOs. >> >>Signed-off-by: Rick Jones >>Signed-off-by: Lamont Jones > > > Thanks for doing this work Rick. > > But as John Heffner and I both mentioned, it's pretty clear we should > do this as a routing metric. Both for handling realistic scenerios > where the sysctl doesn't work, and to help prevent misuse (example: > someone decides that it would be _totally_ _awesome_ for "Carrier > Grade Linux" to set this to 3 seconds by default in /etc/sysctl.conf > and crap like that). If nothing else it was worth the practice :) I'll be happy with either mechanism, just wasn't sure if the jury was still out on whether making it a routing metric was really necessary. I can see where it would be goodness if one had separate paths out of a system, one with the highly variable RTT and one with non-trivial loss rates, just that thusfar I've not come across any :) I've only seen one path with high RTT variability and the other path with trivial loss rates. Also, not surprisingly, the folks for whom I'm doing this are a triffle "anxious" so I figured that simplicity was worthwhile. Particularly if it was going to be the case those folks were going to be asking for back-ports. Anyhow, I'll try grubbing around the source code (already doing that to see about writing a pet tcp cong module) but if pointers to the likely relevant files were available I could try to help thrash-out the routing metric version. Like I said the consumers of this are a triffle well, "anxious" :) rick