[Iccrg] Explicit feedback

[John Leslie]

Doan B. Hoang <dhoang at it.uts.edu.au> wrote:
> 
> One should consider per interface queue length rather than numerous
> individual queues.

   One can reasonably consider queue length for as many queues as you
maintain; but it's hard to justify maintaining very many queues.

> I believe that it is better to deal with congestion control at two 
> levels (at least): the network level and the transport level.
> The transport level never works well if the network level cannot handle 
> congestion properly.

   Actually, transport level works amazingly well under those conditions;
but it seems obvious that _some_ limits to network congestion are needed.

> For example, TCP traffic cannot compete with UDP traffic since there is 
> no cooperation between sources.

   There is a very important issue here; but that's not the right way to
state it: TCP traffic suffers badly when it competes with traffic which
does not practice "TCP-friendly congestion control".

> Furthermore, per-TCP flow congestion control is not scalable (ECN
> modification for example).

   Per-flow TCP congestion control is perfectly scalable when practiced
end-to-end. :^)

> Current congestion control at the network level (IP) is implicit and 
> does not contain adequate information for sources to react.

   I'm not sure what you mean. Current congestion control drops packets
at some congestion level, and that's exactly adequate information for
sources to react.

   Our problems are imperfect transmission of congestion information and
sources which fail to react.

> We are considering Explicit Rate, per-Ingress-Egress congestion control 
> for traffic classes (DSCP-like) plus associated admission control.

   This sounds interesting. Any mechanism which propagates congestion
information back towards the source holds promise of improving the
adaptation to congestion.

> This information will be used by the transport protocol to handle its
> own flow.

   Again, I'm not sure what you mean...

--
John Leslie <john at jlc.net>