From mboxrd@z Thu Jan 1 00:00:00 1970 From: Hagen Paul Pfeifer Subject: Re: [PATCH] sch_red: fix red_change Date: Wed, 7 Dec 2011 23:57:48 +0100 Message-ID: <20111207225747.GA4758@nuttenaction> References: <1322684749.2602.3.camel@edumazet-laptop> <20111130143642.7130aa2b@nehalam.linuxnetplumber.net> <1322773594.2750.28.camel@edumazet-laptop> <1322776643.2750.45.camel@edumazet-laptop> Mime-Version: 1.0 Content-Type: text/plain; charset=iso-8859-1 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: Eric Dumazet , Dave Taht , Stephen Hemminger , Thomas Graf , netdev , Jim Gettys To: Ilpo =?iso-8859-1?Q?J=E4rvinen?= Return-path: Received: from alternativer.internetendpunkt.de ([88.198.24.89]:44454 "EHLO geheimer.internetendpunkt.de" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1755730Ab1LGW5u (ORCPT ); Wed, 7 Dec 2011 17:57:50 -0500 Content-Disposition: inline In-Reply-To: Sender: netdev-owner@vger.kernel.org List-ID: * Ilpo J=E4rvinen | 2011-12-05 13:42:44 [+0200]: >I disagree. If there's any slow starting flow that alone can fill the=20 >bottleneck, anything significantly larger than RTT just harms. RED is=20 >just "too slow" if you follow the recommended parametrization.. > >In a core router you can probably get away with multiple RTTs, but nea= r=20 >edge that is a grave mistake due to how slow-start behaves. With avera= ge=20 >based on many RTTs, RED still estimates that the link has low load whi= le=20 >congestion has escalated to higher dimensions due to slow start. As a=20 >result, RED graciously falls back to tail-drop once the physical queue= =20 >runs out and the flows respond allowing the load to decrease. However,= =20 >finally RED reaches a state where it starts to "pro-actively" react to= an=20 >"incipient congestion"?!? :-/ =3D> Problem is made worse by those extr= a=20 >drops/marks happening too late. But then one question Ilpo: drive IW10 or IW14 the behavior even worse? Especially if n connections start almost simultanously? You did some an= alysis on this topic. HGN