Maybe from a practical implementation perspective in this one instance, but never the less the theoretical discussion on what is technically achievable and the way to do it is quite interesting to some readers. Believe it or not, I actually went and read both of Joe's earlier-referenced RFCs right through from start to end!
To throw something else into the mix with this discussion (apologies if I'm hijacking your thread here Matthew), here's a scenario:
A company is looking to host a VPS somewhere 'in the cloud'. They don't need any IPv4 address space, but they do need a couple of IPv6 addresses (maybe a prefix). I'll just repeat that bit to clarify - the VPS doesn't need any v4 connectivity at all, it will communicate via v6 only!
They would like their VPS to be connected via dual upstreams each with a physically diverse path into the datacenter. The DC would of course have all the usual power backup bells and whistles etc. For latency reasons, the VPS needs to be hosted on hardware that is housed within a DC physically located in New Zealand. As a side note, it will communicate solely with (v6) client IPs who are on NZ UFB links. Ie the clients are only ever on fibre, and are only to be connected to ISPs offering v6 over their UFB links (you know who you are, you sweeties ;-)
The question: what NZ 'cloud hosting'(*) providers would currently offer a service(**) suitable for this VPS?
Second question: if the server operator decided to 'do it themselves' instead, what kind of list of fake and 'possible' devices would they have to come up with for a minimum v6 allocation from APNIC? (or maybe that's better a topic for another talk from Dean at conference ;) (***)
Pete
(*) They don't have to actually use this term in their marketing to qualify! In fact, they may perhaps even score better if they don't.
(**) I'm happy to take replies off-list if it's not good etiquette to discuss such lists on-list or if you're shy
(***) I'm semi-serious here too though. It is plausible for small v6 operators to PI allocations and look to multi-home themselves like has been done with v4 in the past? Or is that undesirable with v6 from the perspective of the global routing tables? (yes I've already registered for the BGP training workshop pre-conference ;)
On 8/11/2013, at 3:26 PM, "Dobbins, Roland"
It seems there's a certain reluctance to look at moving onto someone else's patch and onto someone else's hardware; it might be worth determining whether that reluctance is really warranted, in this day and age.