The DNS is a node identifier, but it currently can't be used by the
transport layer as a connection endpoint identifier.
And neither can a interface address, at least not in the same way. As you
noted, one could use a MIP Home Address as a CEPID, because it is invariant
across interface failures, routing changes, and motion.
There are have also been some experimental proposals (viz. Cheriton's
TRIAD work) to use DNS names for routing.
Yes. If you believe that names should be in the application layer and
addresses should be in the network layer, moving names down is as
problematic as moving addresses up, for the same reason.
