AW: [e2e] Queue size of routers
David P. Reed
dpreed at reed.com
Fri Jan 17 17:38:02 PST 2003
I suspect it's more complicated than this, even. Nobody should wants huge
buffering *in* the network, because that blows the control algorithms in
TCP up. TCP-friendly congestion control needs packet drops (real or RED,
or ECN-simulated) in order to congestion control effectively. Delaying
congestion control with too large buffers is a bad idea, because it causes
spreading traffic jams in the backbone.
When sum(ingress) > egress, it's important to let the contending sources
know quickly, so they backoff quickly. A big buffer will slow the time to
first dropped packet, thus delaying adaptation at the sources.
And of course the other problem with big buffers is that it really kills
traffic with real-time constraints.
The ideal state of the network is that every egress queue has between 0 and
1 packets in it nearly all the time. If queues start having 10 or 20
packets in them as a stable state, that's really bad.
At 03:41 PM 1/17/2003 -0500, Raghurama 'REDDY' wrote:
>Here is my understanding -
>BW*Delay is an end-2-end layer-3 concept (for example in TCP) that need
>to have that much buffering, primarily in the end hosts, in order to do
>Buffering in backbone routers is necessary for a different reason. They
>may have multiple igress and multiple egress interfaces. At any point
>in time it is possible to have:
> sum(input-rates) > capacity of the output link
>So backbone routers also need buffering. Even though this does not
>explcitily have anything to do with BW*Delay product, it has to be
>proportional to BW in order to proive buffering for a certain length of
>From: SMTP%"michael.welzl at uibk.ac.at" "Michael Welzl" 17-JAN-2003
>To: minshall at acm.org, avg at kotovnik.com
>CC: end2end-interest at postel.org
>Subj: AW: [e2e] Queue size of routers
> > -----Ursprungliche Nachricht-----
> > Von: end2end-interest-admin at postel.org
> > [mailto:end2end-interest-admin at postel.org]Im Auftrag von Greg Minshall
> > Gesendet: Freitag, 17. Janner 2003 18:38
> > An: Vadim Antonov
> > Cc: end2end-interest at postel.org
> > Betreff: Re: [e2e] Queue size of routers
> > > Routers in real backbones have the delay*bw of buffer space.
> > good!
>I'm serious - I know that a delay*bw queue length is just
>right if, for example, you suddenly fill the capacity of a
>dumbbell bottleneck in a simulation with new flows and
>don't want some of the initial packets to be dropped,
>thereby eliminating a potential traffic phase effect. But
>is that a good choice for a backbone router?
>The LA<->Tokyo RTT is quite a bit of delay ... is that
>really reasonable when only a small number of flows may
>show this RTT? Shouldn't backbone routers be more concerned
>with traffic aggregates instead?
>And: is there RED in backbone routers?
More information about the end2end-interest