I'm not sure about the definition of 'unused', just because a network is not visible from the Internet and has no publicly registered ASNs doesn't mean its numbers are not in private use (which AFAIK, was always a legitimate use-case for getting an allocation, and in many ways preferable to reusing RFC1918 space).
Added to that, even if it was seeing only minuscule internal use, the UK government's IT project reputation suggests the renumbering would cost at least as much as the block would sell for, assuming the project would even complete prior to the entire planet properly migrating to v6 and the block losing its value.
If there are no networks defined, as far as RIPE is concerned, it's un-used.
We just got audited for RIPE for exactly this reason, and they made us specify details for all of the networks we use on our allocation to be allowed to keep our address space.
Well, if they were feeling ballsy, they could just declare it unused, break it up, and start assigning it. A lot more network nerds would listen to RIPE than to some random UK pension bureaucracy.
Under all circumstances: I'm sure we've all been faced at some stage with the unplanned need to route between two RFC1918 subnets, only to find they collide, in situations as simple as a home VPN connection dialling into the office LAN, up to corporate mergers involving hundreds of thousands of desktops.
A central address registry along with 'public' allocations is the only way to avoid this kind of mess. The fact that public addresses are currently scarce doesn't make having unified addressing any less desirable (just presently impractical).
Sorry, I was imprecise. I meant, under what circumstances today is it the best thing to spend routable IP addresses for machines that aren't exposed directly to the Internet? I'm getting directly at the practicality of these schemes. Obviously, if routable IP addresses were easy to get, there would be lots of cases where it would make sense to use them.
I just gave you a reason. Without coordination of private network addressing, those networks essentially speak different protocols, needing horrendous transforms like NAT which only works in specific situations and myriad crap over the application layer (like DNS views) in order to get them to talk.
If anything, today, networks are more likely to end up interconnected than they were in 1994.
> under what circumstances today is it the best thing to spend routable IP addresses for machines that aren't exposed directly to the Internet
Under the circumstance that you have a spare /8 you aren't using for anything.
You could argue that "the best thing" in that case would be giving, selling, or leasing pieces of the /8 to someone who needs the space. But maybe you aren't in the philanthropy business, so giving's out.
And maybe the prices aren't high enough for your taste, so selling or leasing are out. Or maybe there's simply not enough demand on the market to absorb 16m addresses. Or maybe the terms of the allocation agreement say you're not allowed to transfer them, and you're afraid that trying to do so will give the Internet people a justification to give them to someone else.
> if routable IP addresses were easy to get, there would be lots of cases where it would make sense to use them
If your organization has a spare /8, routable IP addresses are easy to get for you, so these cases do make sense, for you.
Added to that, even if it was seeing only minuscule internal use, the UK government's IT project reputation suggests the renumbering would cost at least as much as the block would sell for, assuming the project would even complete prior to the entire planet properly migrating to v6 and the block losing its value.