Neil
You have made some pretty far reaching arguments. We also evaluated
these schemes when we started GMPLS. It is my opinion we chose the
route because we thought the IP-MPLS synergy in routing was strong.
We run OSPF or IS-IS + BGP and we kill the overhead of having two
routing systems with all the ills that go with it. There was also
as John Drake mentioned more flexibility in adoption of capabilities.
We tried once to put these in ATM and were shut down. Only after
several years crank back solutions were introduced with still with
some inflexible resistance.
Next we looked at the signaling protocols and we fought hard to get
flexibility and a connection oriented behavior. It is my opinion that
CR-LDP was the best suited for this and we still use this model
but we have conceded at present to talking to most of the routers
with RSVP-TE. The signaling protocol is really not most of the
work it is the flexibility we were striving for and with the
exception of refresh the two protocols stay closely in synch.
As far as addressing everyone seem to agree that we can add
external addresses that have nothing to do with the core addressing
so this is not about addressing.
It is my opinion the best place to do this work has been the IETF
over the last few years and here we use IP protocols. It has been
the most responsive to the work to date. I don't see this suddenly
changing and while you are right about it will look a lot like ATM
when we get through that is because it is a connection oriented problem.
Don