Re: A Transition architecture where neither ISP-NATs nor CPE-NAT64s are neeeded

[Rémi Després <remi.despres@free.fr>]

Brian E Carpenter wrote :

> I think this  is a very interesting approach. We could summarise in
> two phrases, I think:
>
> 1. Tunnel to NAT 2. Borrow Address/Port
What about:
- Tunnel to Address/Port server
- NATv4 before tunnel if the client is v4-only

> However, it doesn't solve the case of a genuine "unistack" IPv6 host,
Right.

> and we have to decide whether that case is of real concern.
I agree.
My view on this has changed from "We should do something" to "We can
specify simple enough dual-stack solutions. The need is no longer
clear".

Some may judge however that there is such a need, and may wish to
standardize for this a NAT64, with all its ALG iplications.
At least, their approach should not preclude simpler and more
powerful solutions for more basic transition needs.


> I was also wondering whether the work in SOFTWIRE already covers at
> least part of this solution?

You are right. The draft should probably have discussed it.
- The Softwire solution is however such that dual-stack hosts, or CPE
routers, can borrow *addresses*, NOT *address-port combinations*. This
IMO is a major functional difference (public-IPv4 address space is used
much less efficiently than with NATs or with APBP).
- It imposes that all connections of a client host traverse the same
softwire server (less scalability).
- It implies substantially more complexity in dual stack hosts.

Regards.

Rémi




> On 2008-04-08 01:23, Rémi Després wrote:

> > http://www.ietf.org/internet-drafts/draft-despres-v6ops-apbp-00.txt