On Feb 25, 2013, at 19:35 , Lorenzo Colitti <lore...@google.com> wrote: > > However, the moment you try to use more /64s internally than you have > externally, stateless NPTv6 doesn't work any more, right?
Correct. But remember: I *never* write NAT66 when what I mean is NPTv6. I really did mean NAT66 and not NPTv6. In this scenario, we would number HOMENET domains with 16 bits of ULA subnet identifier and 64 bits of interface identifier for hosts on each subnet. Then we would NAT66 the ULA /48 prefix into whatever addresses are in the pool of longer prefixes the service provider gives us because they won't give us a /48 at an acceptable price. We would not use NPTv6, as that doesn't work. We would then use the PCP protocol to control NAT66 mappings just the same as we do today with NAT44 mappings, but the good news is that PCP explicitly supports that form of IPv6/NAT so it's all good. Thank you PCP working group. Yes, this breaks referral in IPv6 even more than it's already broken, but I'm thinking service providers will be happy with that overall. If I were building an IPv6 home gateway to support routed home networks today, this is how I would feel compelled to do it. -- james woodyatt <j...@apple.com> core os networking _______________________________________________ homenet mailing list homenet@ietf.org https://www.ietf.org/mailman/listinfo/homenet
