"E.B. Dreger" <eddy+public+spam@noc.everquick.net> asks:
Although I generally agree, how does one keep QoS out of the core for CBR
and jitter-sensitive applications?
Magic technology called SONET. In general, applications that really
need CBR/ultra-low-jitter should stay on TDM, because it's the cheapest
approach. Meanwhile applications which can benefit from the
statistical multiplexing gain of the imperfect Internet (with
occasional bursty loss and jitter) can migrate over to cheaper
IP based networks.
There are probably not very many applications which really need
CBR/ultra-low-jitter: most that use TDM networks today work reasonably
well across today's imperfect Internet with no QoS assistance
whatsoever.
As to the core: if there's no queue, there's no opportunity for
(work conserving) fancy queueing. Does it pay to do work non-conserving
fancy queueing? Does it pay to do fancy queueing on transients?
Obviously if you don't, you can't perfectly simulate a TDM-based
network, and thus can only make statistical promises about bit rate and
jitter.
I believe that for perfection it's cheaper to maintain a real TDM
network than it is to use fancy queueing and fancier signalling
to avoid and manage transient congestion in an IP based one.
Sean.