> > > To allow the scenario mentioned above to work, hosts would > > > have to communicate using their link-local addresses. > > > > > > This seems like a reasonable suggestion, any objections? > > > > yes. I strenously object to any expectation that ordinary > > apps should be able to use link-local addresses. > > > > => I should clarify that this is the last resort, > i.e. if no other addresses exist. So first the link > should be configured with a single prefix, if it were not, > there is nothing left but link-locals.
it would be far better to have the hosts on the link elect a PUPI as a prefix to use on that link than to expect apps to use link-locals, because the transitions between global prefixes and PUPIs are far less disruptive than the transitions between globals and link-locals. in general the transitions between configured and unconfigured networks, and between externally-connected and isolated networks, haven't been worked out yet. (or if they have, I've missed the document that describes this...) they're neither simple nor easy to solve, and we must not treat them as if they are simple. -------------------------------------------------------------------- IETF IPv6 working group mailing list [EMAIL PROTECTED] Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------
