From mboxrd@z Thu Jan 1 00:00:00 1970 From: Eric Dumazet Subject: Re: [PATCH] net_sched: sfq: add optional RED on top of SFQ Date: Fri, 06 Jan 2012 18:25:59 +0100 Message-ID: <1325870759.2911.37.camel@edumazet-laptop> References: <1325766316.2415.18.camel@edumazet-HP-Compaq-6005-Pro-SFF-PC> <1325770777.2415.35.camel@edumazet-HP-Compaq-6005-Pro-SFF-PC> <1325867504.2911.23.camel@edumazet-laptop> <20120106090929.009a78d6@nehalam.linuxnetplumber.net> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: David Miller , netdev , Dave Taht To: Stephen Hemminger Return-path: Received: from mail-ww0-f44.google.com ([74.125.82.44]:45377 "EHLO mail-ww0-f44.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752395Ab2AFR0E (ORCPT ); Fri, 6 Jan 2012 12:26:04 -0500 Received: by wgbdr13 with SMTP id dr13so1966110wgb.1 for ; Fri, 06 Jan 2012 09:26:02 -0800 (PST) In-Reply-To: <20120106090929.009a78d6@nehalam.linuxnetplumber.net> Sender: netdev-owner@vger.kernel.org List-ID: Le vendredi 06 janvier 2012 =C3=A0 09:09 -0800, Stephen Hemminger a =C3= =A9crit : > Since SFQ is classful, I don't see what this buys over just putting > a red qdisc under each SFQ class? >=20 You meant RED is classful, not SFQ ? Idea is to use RED on each flow, not on global SFQ, since as you pointe= d out, you already can do that right now (and it has no interest) You can have this right now using QFQ, and for each QFQ class, attach a RED qdisc, but memory cost and setup time are crazy. Please note that RED cant perform head mark or drop. (unless we add new qdisc ops for this) We can do it from SFQ because we do our proper queue management.