From mboxrd@z Thu Jan 1 00:00:00 1970 From: David Miller Subject: Re: [PATCH] tcp: don't allow unfair congestion control to be built without warning Date: Fri, 27 Oct 2006 15:12:38 -0700 (PDT) Message-ID: <20061027.151238.59655130.davem@davemloft.net> References: <20061027142402.07a47f3f@freekitty> <20061027.143701.85411558.davem@davemloft.net> <20061027145913.04dc0e5f@dxpl.pdx.osdl.net> Mime-Version: 1.0 Content-Type: Text/Plain; charset=us-ascii Content-Transfer-Encoding: 7bit Cc: hagen@jauu.net, jheffner@psc.edu, netdev@vger.kernel.org Return-path: Received: from dsl027-180-168.sfo1.dsl.speakeasy.net ([216.27.180.168]:38325 "EHLO sunset.davemloft.net") by vger.kernel.org with ESMTP id S1750764AbWJ0WMh (ORCPT ); Fri, 27 Oct 2006 18:12:37 -0400 To: shemminger@osdl.org In-Reply-To: <20061027145913.04dc0e5f@dxpl.pdx.osdl.net> Sender: netdev-owner@vger.kernel.org List-Id: netdev.vger.kernel.org From: Stephen Hemminger Date: Fri, 27 Oct 2006 14:59:13 -0700 > On Fri, 27 Oct 2006 14:37:01 -0700 (PDT) > David Miller wrote: > > > From: Stephen Hemminger > > Date: Fri, 27 Oct 2006 14:24:02 -0700 > > > > > Only some (very few) have any bad consequences. So the typical > > > distribution should be able to switch with most available for everyone, > > > and only a few needing special privileges. > > > > I would strongly disagree as we've had several OOPS'er class bugs in > > the less frequently used algorithms. > > > > Then tag those as restricted. Why should we keep app's away from > the simple ones. You can't predict bugs, but what you can do is know that the lesser used algorithms are by definition less tested and therefore more likely to have bugs. Everything except the default and Reno are lesser used. Safe by default, there is no other choice. You fail to respond to THAT part of my email. That's the important point. Let me reiterate: > It's bad enough that people are all over us for the default algorithm > we have choosen, so it'd be extremely irresponsible and even worse if > we allowed users to select any of the other "research" algorithms for > their TCP connections by default just because those modules happened > to be configured into the kernel. > > This userspace convenience argument holds zero water. > > Provide a way for the administrator to control the situation fully, > and choose a sane default which errs on the side of caution for the > sake of internet stability. Please reread this and consider why it's important.