Re: [v6ops] Scope of Unique Local IPv6 Unicast Addresses (Fwd: New Version Notification for draft-gont-6man-ipv6-ula-scope-00.txt)

[Ted Lemon <[e-mail]>]

On Jan 6, 2021, at 2:30 AM, Christopher Morrow <[e-mail]> wrote:
> option 4, deprecate ULA.
> the best option (tm).

If that were an option, we wouldn’t be having this argument—nobody would care about ULAs. ULAs are a good idea. The terminology around them needs work, that’s all. 

For example, we use ULAs in the HomePod Mini to route between adjacent network links where IPv6 GUA delegation isn’t available. The ULA never winds up in the global routing topology. The Mini chooses it using a secure RNG, so the likelihood of collision is vanishingly small. ULAs are _much_ more flexible than RFC1918 addresses, simply by virtue of the process by which the /48 prefix is chosen.

I would have major wibbles about using RFC1918 addresses in the Mini the way we currently use ULAs, because we’d have (at best!) eight bits of randomness, and a strong likelihood of collisions with competing private network uses of the 10.0/8 space. Because ULA is specific about each prefix being a /48, and because a /48 is most likely enough for most use cases, the worries about this sort of collision are nonexistent: nobody is going to allocate the whole ULA space to a single site, and if they do, we can legitimately say that they are at fault for things not working.

We can’t and shouldn’t deprecate ULAs. I think clarifying what the name means makes sense, though, and perhaps the term should be CUA (collision-unlikely address) prefix rather than ULA prefix.