Re: new draft on IPv6 CPE router available for review

[Shin Miyakawa <miyakawa@nttv6.jp>]

David,

> Ole raised the issue that a router would (almost by definition) need  
> to implement a weak or semi-strong model, and I would support this  
> view. I do believe the CPE must support the sending of IP datagrams to  
> Internet hosts (for ICMPv6, management, etc) but see an option of  
> using a loopback interface for this purpose (from the subscriber's  
> delegated prefix). 

Please note well that, if you'd like to assign a global IPv6 unicast address
to the LoopBack interface,
because loopback is the diffrent from LAN links behind the CPE,
we have to assign at least two /64 prefies to the CPE
(one /64 for LAN, one /64 for loopback, assuming that a /64 is the minimal unit 
of prefix delegation).

> This allows a single subscriber to be represented  
> by one prefix that may be operationally beneficial, or provide a more  
> scalable BNG architecture.

So, this consumes more than 
"one Global address for the WAN interface with one /64 for LAN" (as
I called this as "1+/64" or for some reason, "1+/64s") model.

How do you think about this fact ?

Or forget about the loopback interface and use the CPE's LAN interface in
the weak host model case ?

Best wishes,


Shin Miyakawa