From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from bifrost.lang.hm (mail.lang.hm [64.81.33.126]) (using TLSv1 with cipher DHE-RSA-AES256-SHA (256/256 bits)) (Client did not present a certificate) by huchra.bufferbloat.net (Postfix) with ESMTPS id 12F5820050B for ; Wed, 24 Sep 2014 21:16:47 -0700 (PDT) Received: from asgard.lang.hm (asgard.lang.hm [10.0.0.100]) by bifrost.lang.hm (8.13.4/8.13.4/Debian-3) with ESMTP id s8P4GiWx007621; Wed, 24 Sep 2014 21:16:44 -0700 Date: Wed, 24 Sep 2014 21:16:44 -0700 (PDT) From: David Lang X-X-Sender: dlang@asgard.lang.hm To: Mikael Abrahamsson In-Reply-To: Message-ID: References: User-Agent: Alpine 2.02 (DEB 1266 2009-07-14) MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII; format=flowed Cc: bloat Subject: Re: [Bloat] I feel an urge to update this X-BeenThere: bloat@lists.bufferbloat.net X-Mailman-Version: 2.1.13 Precedence: list List-Id: General list for discussing Bufferbloat List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Thu, 25 Sep 2014 04:17:16 -0000 On Tue, 23 Sep 2014, Mikael Abrahamsson wrote: > On Sat, 20 Sep 2014, Dave Taht wrote: > >> http://tools.ietf.org/html/draft-gettys-iw10-considered-harmful-00 >> >> I just got some iw10 results on a t1 line... ugh. anyone want a stab at it? > > Is it at all possible to create a solution that works well for all cases? > > I pitched a proposal at IETF75 and then again a year ago in TCP/TCPM WGs that > it might be good if the connection manager could hint the TCP stack, or the > TCP stack could determine itself by heuristics, that its network connection > were of a few different degrees of a few criteria which might be speed, loss, > jitter etc. I perhaps even could tell my connection manager that when > connected to my home wired or wifi network, "everything" is reachable via 50 > + megabit/s connectivity. If I then speak to a 10GE connected server, there > should be no problem for it to do IW10. Basically, my thinking is to have > something similar to "MSS" but when it comes to connectivity. > > There was no interest anywhere in this, everybody wanted for each connection > to be living in its own universe with little prior knowledge about what's > been going on before it. The problem is that you don't know what the connectivity is going to be. (unless you are connecting to the same IP as an existing connection). Your first few hops are fairly predictable, but after that you have no idea if you are going to be connecting to a server on a low bandwidth link, one behind a very congested router, or one with better connectivity than you have. > I just don't see how we can make things work well when we try to create TCP > to handle everything from 1500ms to 0.1ms of RTT, and everything from > 19200bps to 100GE in speed. Actually, at this point, we know exactly what would need to be done to make TCP work across this entire range (as long as the connection speed doesn't vary too rapidly), the problem is getting the hardware manufacturers on board to actually deploy it. using fq_codel on every bottleneck link will make TCP work pretty well across that entire range of connectivity David Lang