Brian E Carpenter wrote:
marcelo bagnulo braun wrote:...The endpoints are the only place that can or should know about whether they change.This is a purist view of the Internet we used to have when RFC 1958 was written... the current reality (RFC 2775) is that all these problems call for soft state solutions, that can be garbage collected as Marcelo says, and recreated if they are prematurely lost.
Soft state elsewhere is an optimization that relies on hard state at the endpoints for its recovery and recreation anyway. I.e., I don't see yours and Marcelo's views as inconsistent.
(The tragedy of NATs is that they can't do the second bit reliably, and multi6 needs to avoid that problem.) Brian