Kimmo Suominen wrote: I've used IPv4 space in the past to connect to extranet providers using unique addresses, without connecting to the Internet. Is such use not allowed in the IPv6 world? One must promise to advertise the addresses to the Internet to get an allocation?

Although this is blurry (no explicit requirement) we can say that as of today, yes. But the space LIRs get is big enough to provide both Internet and extranet services. I don't see any extranet-only provider getting address space now though, and for multiple reasons.

Gert Doering wrote: I wouldn't go so far as to say "it's not allowed".

Me neither. Actually, I don't see why it should be forbidden, as long as one obtains the address space, which is the issue here.

When the policy was made, people were still suggesting the use of site-local addresses for "non-global" usage. Site-locals seem to be dead, so there is a hole in the policies right now. Passing on the question from the registry point of view to the IETF people (Michael & co): what are your recommendations how this can be addressed (in the double sense)?

[disclaimer: I do not represent the views of the IETF] This is a complex answer. First, let's not leave site-locals for dead yet. Technically, we do have site-locals using the "full usage" model, as defined by RFC 3513 that was just published. The current situation is that there is a "consensus" to deprecate them, which has pissed so many people that appeals are lined up for the next 2 years already (take-a-number if you want to appeal). There is no actual text to remove site-locals and it is expected that any text that would attempt to do that will be stalled and never go forward. In short: technically speaking we currently do have site-locals with an RFC in the standards track and I don't expect any change any time soon. That being said, the reason we got into this deadlock is that site-locals as currently defined do not please many people. If there is change in leadership within the IETF and work on site-locals is resumed (instead of trying to get "my way or no way") it is expected that site-locals will be restricted to a model that prohibits communication between sites. So, in any case I would not use site-locals for communication between sites. There are ideas floating around to make them globally unique, but this is for the purpose of avoiding renumbering when merging sites and not to provide site-to-site communication. Global addresses are required for that purpose, whether or not they are publicly routed or not. There are several proposals to provide PI-like addresses that are moving forward though. Michel.