> does the rendezvous location really have to be the "original topological > location" of the host or is that just how folks started thinking about it? > > and given that the rendezvous location has to be somewhere in the network, > how can we get around the problem that that location might become > unreachable no matter where it's located? I mean, you could replicate > the information to multiple sites and use anycast to find a location, but > would this fundamentally change the mobile IP protocol?
How would you do this such that you did not further aggravate route scaling problems? If you do it with mobile-ip and anycast, would that would require such mobile addresses having either additional IGP or (worse) BGP paths? On the other hand, if you have the rendevous be a database that maps some sort of a name to an IP address you do not perturb the routing system and you impose the overhead on those who are mobile, and their agents. It requires a model without TTLs but initialization and active notification. Whether this can be done is questionable and a valid topic of research (a bunch of us have been questioning for a while ;-) Eliot