Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 9 Jul 2004 01:41:07 -0700
From:      Luigi Rizzo <rizzo@icir.org>
To:        Ian FREISLICH <if@hetzner.co.za>
Cc:        "Thomas S. Crum - 1WISP, Inc." <tscrum@1wisp.com>
Subject:   Re: Dummynet Queue Weighting
Message-ID:  <20040709014107.A35991@xorpc.icir.org>
In-Reply-To: <E1BiqbF-000DWn-00@hetzner.co.za>; from if@hetzner.co.za on Fri, Jul 09, 2004 at 10:19:17AM %2B0200
References:  <louie@TransSys.COM> <E1BiqbF-000DWn-00@hetzner.co.za>

next in thread | previous in thread | raw e-mail | index | archive | help
On Fri, Jul 09, 2004 at 10:19:17AM +0200, Ian FREISLICH wrote:
...
> > > But, what if A&B(1) had no traffic? It is my understanding that
> > > queue 2 and 3 would still only get 5/16 and 1/16 of the pipe
> > > regardless. In this example, 3/8 or 375Kb/s total. Or would 2 and
> > > 3 share the whole pipe if queue 1 is inactive, which would make my
> > > questions moot?
> >
> > The answer to your question is that idle queue do not consume capacity
> > on the pipe they are associated with.  I have queue with weights 100
> > (for VoIP), 20 (for interactive SSH, NTP) and 1 (everything else) and
> > the "everything else" traffic can use the full capacity of the pipe
> > with the other queues are idle.
> 
> This raises another question: how do the idle queues get shared?

the only thing that is shared is the total pipe's capacity,
and it is shared by non-idle queues in proportion to their
weights. That's as simple as that. No special cases.

There is a copious literature on Proportional Share
algorithms, if you google for WF2Q+ (which is the algorithm
implemented in dummynet) you should come up with a lot of
papers to answer your doubts.
We are finishing up a tutorial paper on the topic for which i will
post a URL in a week or two when it is ready.

cheers
luigi



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20040709014107.A35991>