lachlan.andrew at gmail.com
Wed Aug 8 02:30:31 BST 2007
On 07/08/07, dimitri papadimitriou <dpapadimitriou at psg.com> wrote:
> Bob Briscoe wrote:
> > - what's the minimum information intensity of congestion notification
> > (More than 1-bit per packet? Is sensing congestion delay variation a
> > sufficient alternative?)
> yes this is imho an important point. would deserve additional discussion
> on this list. i would distinguish intensity from encoding granularity
> and significance.
Yes, you raised this point in October last year too. At the time, I
think the consensus was that we need to distinguish between "1
information-theoretic bit per packet" and "1 protocol bit per packet",
which generally gives much less information. That may be related to
Dimitri's point about separating the signalling rate from the
semantics of the signal.
>From a control theory point of view, there's recently been a lot of
work on the minimum information-theoretic signal rate required to
"stabilise" a system (e.g,
<http://www.ee.unimelb.edu.au/staff/gnair/NairPIEEE.pdf>). I'm not
sure how much of that relates to congestion control, where
undershooting the rate is much preferable to overshooting.
To decide how much feed back is "needed", we need to decide what
performance measures to use, and what performance level is "needed" --
e.g. do we need unused bandwidth to be taken up again within x
RTTs/seconds/minutes of being released by another flow? Do we need
overload to abate within x RTTs?
Lachlan Andrew Dept of Computer Science, Caltech
1200 E California Blvd, Mail Code 256-80, Pasadena CA 91125, USA
Phone: +1 (626) 395-8820 Fax: +1 (626) 568-3603
More information about the Iccrg