Many folks would consider that a benefit. Hosts within an
enterprise
need to be managed in a scalable manner, which effectively
necessitates the proxy approach. This is the same driver that makes
proxy mode for shim6 a pragmatic necessity. Along with that proxy
mode, it would seem that this would relieve the burden on the
host of
learning and retaining all of the addresses in its bunch. This
would
actually simplify deployment.
The alternative approach is that the necessary policy and preference
table is automatically distributed to all hosts in the site from
a central point; that would of course require overhead in every host
(but RFC 3484 already assumes such overhead).
Right. And in the specific case of Six/One, you may not even need the
automatic policy/preference distribution.
Given Six/One support on hosts, all the network needs to do to
enforce a
particular provider for a given connection is to write that provider's
routing prefix in the packets of the connection. The end hosts will
recognize the provider change and adopt it -- which also allows for
some
host-internal adaption at the transport and application level for
ongoing connections.