6rd update & multi-subnet support to be possible for v6 /32-only ISPs

["Rémi Després" <remi.f.despres@gmail.com>]

Brian Carpenter wrote :
> On 2008-03-12 07:44, james woodyatt wrote: ...
>>
>> The draft defines a new DHCPv4 option used by 6RD ISP to configure
>> 6RD-enabled CPE, i.e. hosts or dual-stack IPv4/IPv6 routers with
>> 6RD-tunneling functions.  I very much doubt that getting vendors of
>>  consumer retail CPE, e.g. host operating systems, consumer
>> routers, etc., to adopt this DHCP client feature will be easy.  For
>> that reason, I don't expect 6RD to be deployable except with
>> provider-provisioned CPE.
>
> And that is unfortunate, because it will lock customers into a /64
> world until the ISP moves on from 6RD to a real IPv6 deployment,
> which as just noted on another thread, is a problem in itself. I
> sympathise with the goal of rapid deployment, of course, as long as
> it doesn't become a dead end.

1. Two reasons why this conclusion is IMO pessimistic:
 (A) If a customer site has several v4 addresses, it has by the same
token several v6 subnets.
 (B) If an ISP starts with a /32 prefix, and later gets a shorter one
(this happens to be the case for FREE, which obtained a /28 after its
initial /32), its customers can, even with the current 6rd proposal,
get several IPv6 subnets.

2. The goal of 6rd is certainely not not do delay deployment of more
native v6 prefixes. It is intended to provide for rapid deployment of
simple and safe IPv6 addresses to customers of ISPs that, for various
reasons, are still reluctant to modify the core of their
infrastructures.

3. I agree with James that the easiest and most likely use of 6rd, at
least initially, is where router CPEs are controlled by iSPs (it was
first devised for them). However, some ISPs MAY chose to extend this
way to provide IPv6 addresses to CPEs that are  directly connected
hosts (a positive perspective). But for this, they MUST know which
DHCPv4 option to implement to make this practical.
This is the why this option should be finalized asap, especially for
implementations that are long to make.

4. DESPITE REMARKS 1 and 2, it appears that a compromise between
simplicity and functionnality different from that of the current
proposal is more amenable to a consensus.
It should allow for several subnets per customer site even if ISPs,
(1) only get /32 v6 prefixes from their RIRs, (2) provide only one v4
address per site.
Tony Hain proposed some ideas in this direction.
I  work with him on it.

Regards.

Rémi

Note: I could not participate in discussions since tuesday morning
when my mac crashed :-) .
I only have this new provisional email to communicate while here in
Philadelphia.