From mboxrd@z Thu Jan 1 00:00:00 1970 From: David Miller Subject: Re: 2.6.20.7 TCP cubic (and bic) initial slow start way too slow? Date: Wed, 13 Jun 2007 00:56:42 -0700 (PDT) Message-ID: <20070613.005642.85685043.davem@davemloft.net> References: <20070612.151258.120619253.davem@davemloft.net> <20070612171603.47736395@localhost> <20070612233814.4b04db2f.billfink@mindspring.com> Mime-Version: 1.0 Content-Type: Text/Plain; charset=us-ascii Content-Transfer-Encoding: 7bit Cc: shemminger@linux-foundation.org, sangtae.ha@gmail.com, rhee@eos.ncsu.edu, rhee@ncsu.edu, netdev@vger.kernel.org To: billfink@mindspring.com Return-path: Received: from 74-93-104-97-Washington.hfc.comcastbusiness.net ([74.93.104.97]:60459 "EHLO sunset.davemloft.net" rhost-flags-OK-FAIL-OK-OK) by vger.kernel.org with ESMTP id S1756059AbXFMH4Z (ORCPT ); Wed, 13 Jun 2007 03:56:25 -0400 In-Reply-To: <20070612233814.4b04db2f.billfink@mindspring.com> Sender: netdev-owner@vger.kernel.org List-Id: netdev.vger.kernel.org From: Bill Fink Date: Tue, 12 Jun 2007 23:38:14 -0400 > If there was a benefit, perhaps it would be useful to have a > per-route option for setting the initial_ssthresh. We have this per-route setting already, BIC and CUBIC just override it with their local initial_ssthresh value when that is non-zero. I wonder how many folks get tripped up by that totally unexpected behavior. What's the point of the route metric we've had all of these years if the congestion control algorithm is just going to totally ignore it? That should never be the default behavior. > OTOH just setting it to zero as a default should also be fine as > that's the standard Reno behavior. This is what I'm going to do. Injong-ssi and Sangtae-ssi have had several weeks to give us final guidance in this area, we cannot wait forever.