> | Can we keep 3gpp architecture discussions OFF this list? > > As far as it goes, probably yes - but as long as you keep referring > to peculiarities of it, to justify other decisions, then no...
??? What does this have to do with cellular architectures? We need to consider particulars of the link but that's a totally different thing from the internal architecture of a cellular system which belongs elsewhere. > > The problem here is that there's no spec for IPv6 over those links. > Define one of those first, until that's done, attempting to define > host requirements for hosts connected to such a link is straight out > silly - how can anyone run IPv6 over any random link in any kind of > standardised way, when there's no existing spec on how to do that? > > It is frightening to think that people seem to be > contemplating actual > implementations of this, with no spec at all to base their > implementation > on. This needs to be fixed, and quickly. 3gpp has defined how the IPv6 behaviour over cellular links works with IETF input in terms of 3gpp-advice. It's in a spec called 23.060. So it's simply not true that people don't have a spec. However it is important that we quickly get some form of minimum host requirements spec which today just isn't there to guarantee host compatibility. This could then be used by 3gpp implementers to ensure compatibility but it needs to be done in a short time. In parallel we could do the IPv6 over cellular spec which would give non-3gpp people something to use. It's already been mentioned by Jonne that work has started on this. > > | What we are talking about is an IPv6 host with a point-to-point > | link to a router. They run PPP and the router delegates > | a global prefix to the host. The router doesn't configure > | any address on the delegated prefix and it supports NS/NAs. > > It sounds as if what you're really describing is an > "unnumbered" link. > (From the global point of view anyway) - the delegated > prefix would be > for the host (terminal, phone, ...) to use as it sees fit? Yes. Please read the previous emails on this. > > Is that how it always *must* be done? If so, that's what > the IPv6 over > whatever doc should be saying (along with the appropriate > justification). > > But in any case, any link is going to have the link local > prefix on it, > and on that one, the router is supposed to go and configure > the "any routers" > anycast address, aside from a unicast address for itself > usually - there's > absolutely no reason why it shouldn't - regardless of any > prefix length > that might be used for a global prefix on the link (were > there to be one), > the link local is always a /64 - so there really are plenty > of addresses > there for there to be one for the host/terminal/phone/... > and one for the > router. Not allowing the router to have a link local address on the > link would need very good justification, as it would break > all kinds of > assumptions made everywhere else. Well, if you had read the previous emails you'd have realised that 3gpp is using PPP to guarantee uniqueness of the router and host's link-local. And no, once again, the router does not configure any address on the delegated global prefix. It is a special router. I'm starting to sound like a broken record :( /Karim -------------------------------------------------------------------- IETF IPng Working Group Mailing List IPng Home Page: http://playground.sun.com/ipng FTP archive: ftp://playground.sun.com/pub/ipng Direct all administrative requests to [EMAIL PROTECTED] --------------------------------------------------------------------
