[jonny_storm]

You are quite right that IPv4 is exhausted and unobtainable. Moreover, your intuition about IPv6 is also correct, though I would argue the lack of Internet-scale protocol translation is what actually destroys the IPv6 value proposition.

The limitation you point out is not actually in BGP but is, instead, a matter of convention: ISPs will, generally, accept prefixes longer (smaller) than /24 but will not advertise them to other ISPs, in turn. This convention arose out of the need to mitigate the steady increase of core routing table size--as did CIDR, itself, and VLSM before it.

While not ideal, most organizations use DNS for failover of individual, redundant hosts from one address to another. Of course, you may find this unsuitable for your use case, but if the users of your host's service are "the Internet," then the only alternative is to mediate how those users access the host, either by inserting a proxy or creating redundant connections with a custom client on the host.

If you can provide more detail regarding your specific requirements, I may be able to elaborate further, more bespoke alternatives.

[ThePhysicist]

It’s not true that IPv4 is unobtainable, you can e.g. apply to become a RIPE member & LIR to obtain a /22 subnet of addresses. The cost is in the several 1000 €’s per year if I remember correctly. You can equally well try to reach out to an existing LIR and rent addresses from them of course. Not sure if that corresponds to your definition of “owning” IP addresses but as a LIR you can theoretically hold on to the addresses indefinitely.

More info: https://www.ripe.net/participate/member-support/become-a-mem...

[jonny_storm]

Because RIPE's last /8 is gone, any remaining addresses will come from whatever has been recovered. Under no circumstances should you expect to receive a contiguous /22 through LIR registration. Still, for obtaining routable /24s, the LIR route will likely remain viable for the next year or so.

[jamieweb]

Thanks for the info.

I'm interested in having my own IPs mainly so that I can move servers wherever I want and take the IPs with me. CNAMEs can only go so far when it comes to moving things around without interruption.

It's only a small personal operation though - and I am aware that limitations with hosting providers and ISPs may prevent me from being able to point my own IPs to them.

[jonny_storm]

If you're not satisfied with the low-TTL/round-robin strategy, you would need to build out some private infrastructure to perform routing failover.

A tunnel with private routers between two cloud providers would allow you to advertise private host routes (/32) inside your network while NAT-ing appropriately to the outside. This achieves manual failover (or fast migration), but it can be expanded for dynamic failover, as well.

Let me know if you'd like me to elaborate the above solution. Otherwise, good luck with whatever solution you choose.