From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jarek Poplawski Subject: Re: [PATCH 3/3] pkt_sched: restore multiqueue prio scheduler Date: Mon, 25 Aug 2008 08:25:02 +0000 Message-ID: <20080825082501.GD2633@ff.dom.local> References: <20080825060640.GA2633@ff.dom.local> <20080825.004825.193701182.davem@davemloft.net> <20080825075744.GC2633@ff.dom.local> <20080825.010206.193700650.davem@davemloft.net> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: hadi@cyberus.ca, alexander.duyck@gmail.com, jeffrey.t.kirsher@intel.com, jeff@garzik.org, netdev@vger.kernel.org, alexander.h.duyck@intel.com To: David Miller Return-path: Received: from ey-out-2122.google.com ([74.125.78.24]:28003 "EHLO ey-out-2122.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753450AbYHYIZJ (ORCPT ); Mon, 25 Aug 2008 04:25:09 -0400 Received: by ey-out-2122.google.com with SMTP id 6so161378eyi.37 for ; Mon, 25 Aug 2008 01:25:07 -0700 (PDT) Content-Disposition: inline In-Reply-To: <20080825.010206.193700650.davem@davemloft.net> Sender: netdev-owner@vger.kernel.org List-ID: On Mon, Aug 25, 2008 at 01:02:06AM -0700, David Miller wrote: > From: Jarek Poplawski > Date: Mon, 25 Aug 2008 07:57:44 +0000 > > > On Mon, Aug 25, 2008 at 12:48:25AM -0700, David Miller wrote: > > > From: Jarek Poplawski > > > Date: Mon, 25 Aug 2008 06:06:40 +0000 > > > > > > If we feed packets after the first one to the card, we would not > > > be implementing a FIFO. > > > > Not necessarilly so: if separate flows are hashed to "their" hwqueues, > > a FIFO per flow would be still obeyed. > > What appears on the wire is still going to be similar. > > You have to subsequently ask if it's worth the complexity to do > what you seem to be proposing. > > When a single hardware queue fills up, it's the SAME, semantically, > as when a unary TX queue of a traditional device fills up. > > There is NO on the wire difference. There will be NO performance > difference, because the device will have work to do as by definition > of one TX queue being full there are some packets queued up to > the device. If with unary TX queue we had to fill one bigger queue (or all TX queues) before device stopped a qdisc, and with mq TX it's enough to have one TX filled to effectively stop a qdisc transmit, IMHO there should be a performance difference. Jarek P.