Re: draft-narten-ipv6-3177bis-48boundary-02.txt

[Tim Chown <tjc@ecs.soton.ac.uk>]

On Thu, Jul 13, 2006 at 05:30:16PM +0300, Pekka Savola wrote:
> 
> Per above.  For example, one could explicitly state what the IETF 
> architectural considerations mean in this context.  E.g., this 
> document does not give a warrant to assign addresses beyond /64, like 
> as /128s.

From the home user point of view, as a customer, I would like:

1) At least a /60, given what I have today.
2) A fixed fee (no more 'ah a fixed IP is $10 more sir')
3) A static allocation
4) Moving to a different provider I will get the same size allocation
5) To use ULAs if/when their buglets are ironed out
6) To (when CPEs allow it) source and use IPv6 multicast services

Not having a fixed /48's may lead to differential charging (#2) for 
different offerings and potentially variation between providers(#4),
thus agreement on something is important.  A /56 is ample for actual
subnet use today.

I note that at least one tunnel broker product does not support /56 
size allocations cleanly.   That would need to change.   Users migrating
from 6to4 would have a smaller size allocation, so should form internal
address plans with that in mind for when they 'go native' (if they want
to keep the same topology and subnetting).

I think 3177 could mention static allocatiobns ( I don't see it in there).
This impacts address space consumption by the ISP, but I think is worth
including.

Regarding ULAs, I guess there would be some 'feature' of the CPE that
generates a /48 ULA under fd00::/8.  If we recommend a /56, what should
that CPE do?  Just use address space out of a 40-bit random allocation
with the other 8 bits zero?   ULAs were designed with an 'everyone uses
/48' view in mind.

On multicast, I can still generate unicast prefix based group addresses 
and even an embedded-RP address, as I can do for any /64, so that's OK.

-- 
Tim/::1