[e2e] Can feedback be generated more fast in ECN?

[J. Noel Chiappa]

    > From: "Eric A. Hall" <ehall at ehsco.com>

    > SQ messages are structurally identical to Time Exceeded and
    > Destination Unreachable messages.

Now there's an interesting point, vis-a-vis the point about SQ being too
painful for high-speed routers to generate. AFAIK, those same routers seem
to generate TE messages with no problem. So I gather that they do include
mechanisms to get packets out of the "fast path", and off to a place that
can turn them into ICMP error messages.

Yes, one would expect to see more SQ messages generated than TE, so there
does need to be some rate control on the generation of SQ messages, to
prevent it taking up too many non-forwarding processing resources.


(The point about sending SQ's taking processing power away from processing
routing updates is not really a strong one, since a router being run that
close to the edge, in terms of non-forwarding processing resources, is
already in trouble. Given how cheap processing power is, vis-a-vis the cost
of everything else [transmission capacity, physical installation, etc],
there's no reason to be in any operating mode close to used-up.

In addition, there is always variation in the amount of routing update
traffic, and unless you get a congestion event in synch with a topology
change [and I know the second often leads to the first, but you often do
get congestion without a topology change], you'll normally have spare power
to send SQ's. If you don't have spare power, then don't send the SQ's.)


This is also not, of course, a complete analysis of the cost/benefit case
for including incipient congestion notification in explicit ICMP messages
versus in ACK's. I suspect each has cases in which it is superior, and I
guess the question is does SQ have enough cases in which it is better to
make it worth doing.

No idea, myself...

	Noel