From owner-freebsd-net Sat Nov 13 11:56:10 1999 Delivered-To: freebsd-net@freebsd.org Received: from cs.rice.edu (cs.rice.edu [128.42.1.30]) by hub.freebsd.org (Postfix) with ESMTP id 9E16514E57; Sat, 13 Nov 1999 11:56:07 -0800 (PST) (envelope-from aron@cs.rice.edu) Received: (from aron@localhost) by cs.rice.edu (8.9.0/8.9.0) id NAA23932; Sat, 13 Nov 1999 13:56:03 -0600 (CST) From: Mohit Aron Message-Id: <199911131956.NAA23932@cs.rice.edu> Subject: Re: FreeBSD networking problems To: luigi@info.iet.unipi.it (Luigi Rizzo) Date: Sat, 13 Nov 1999 13:56:03 -0600 (CST) Cc: freebsd-net@freebsd.org, wollman@freebsd.org, jlemon@freebsd.org, julian@freebsd.org, bright@wintelcom.net In-Reply-To: <199911131431.PAA12855@info.iet.unipi.it> from "Luigi Rizzo" at Nov 13, 99 03:31:23 pm X-Mailer: ELM [version 2.4 PL25] MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Sender: owner-freebsd-net@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.org > > what you are saying is "i can blast packets to the interface faster > than it can drain them". Which means that you are causing _congestion_ > on the local interface rather than somewhere in the > middle of the net (and, your burst of 1000 packets might as well go > to your next-hop router where you probably won't have as much queue > space for your flows). No, once the interface has spaced out the packets (because of the b/w of the output link), they don't queue up at the next hop router unless the outgoing b/w is lesser than the incoming one. > so, fix TCP, the problem is there, not elsewhere (possible fix, > not terribly complex: on overflow don't cut the window as done > now, just schedule a call to the output loop after 1/HZ seconds). > I'm not disagreeing with you that TCP should never pump packets at a faster rate than the interface can handle them. Hoever, the mechanism for this is not yet in place. For example, HZ currently is 100 by default - which means you can pump packets spaced by a minimum of 10ms - you can pretty much forget about getting decent b/w at this rate (a 100Mbps link requires 120usec spacing 1500 byte packets for full utilization). You can argue that HZ can be increased, but then the overhead of hardware timer interrupts gets prohibitively large. I've proposed a software technique for achieving low overhead fine-grained timers in my SOSP '99 paper - http://www.cs.rice.edu/~aron/papers/soft-timers.ps.gz. But the default FreeBSD distribution doesn't yet have that of course. - Mohit To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-net" in the body of the message