RE: IP QoS issues

[Karl Auerbach <karl at CaveBear.com>]

> > I agree with Karl that IP QoS is still quite a long way off. It's not as
> > easy as reusing ATM techniques. Instability of routes is bound to maintain
> > QoS difficult and it's not a solved problem.
> 
> Red herring are aggregation, policy and MPLS.
> 
> Also, diff serve is just as bad as ABR/VBR.
> 
> Once you forget all of them, stable routing of qos assured
> traffic (which is different from stable routing of best effort
> traffic) is easy and solved.

That's sort of like saying "when we are all wealthy there will be no
problem of poverty."

Yes, once we have stable routing, many difficulties will go away. 
However, even within an autonomous area, as you suggest in your paper on
QoS routing (with pinned routes), there are, in real-life nets, enough
failures and other sources of instability to make betting on the
permanence of routes rather like betting on a slow horse in a fast race. 

And once one gets out into the big-bad world of inter-ISP routing, the
concern usually isn't whether routes flap, but how fast.

I suspect that you are going to say "route pinning will stop most
flapping".  If you were to say that, I'd answer "I'm not at all sure that
inter-routing-domain pinning will be administratively acceptable to many
operators and that I'm also skeptical that we have the technology to do
that yet between routing domains."

Anyway, I think that we've made our respective opinions known and need not
turn this into a long, drawn-out thread on the IETF mailing list.

		--karl--