Re: Generic anycast addresses...

[Ted Lemon <[e-mail]>]

An update on this, in case anyone’s wondering.   It turns out that PCP does have an anycast address allocated, as do a few other protocols.   There is no explicitly designated anycast prefix, and the anycast address is just a regular global IPv6 address with no scoping rules.   It appears to be the case that the prefix 2001::1//64 has been used for this.

Ironically, I had actually put this in the SRP document a while back, and forgot about it.  This thread was sparked by my attempt to find an anycast prefix with scoping rules, which I failed to do.

FWIW, given that well-known anycast addresses are a current practice, I think it makes sense to have an explicit prefix from which they are allocated.   I do not know what would happen if I sent a PCP anycast—how far it would get before it was dropped due to lack of a route.   But it seems likely that it will escape the site.   This is probably not ideal for a lot of use cases, SRP being one of them.   So it might be worth writing a draft that addresses this and reserves a prefix with site scoping rules that we can safely assume will be followed.   Allocating a prefix out of fc00::/8 would address this problem.

I have no appetite for a general solution to this problem, and I didn’t get the sense from the (rather long) discussion we had on this that anyone else did either, with one exception.   The use case I know we need is site-scoped.

Despite being in Montreal, I will not actually be able to attend 6man because it’s opposite one of the DNS working groups.  If people are interested in talking about this, ping me and we can arrange something.

> On May 29, 2019, at 6:48 PM, Ted Lemon <[e-mail]> wrote:
> 
> I was looking through the IANA registry for anycast addresses with an idea of what I wanted, and was surprised to learn that no such thing exists.   I’m curious if what I want is something that’s already been shot down in flames, or something for which no energy has existed to do.
> 
> Right now it appears that anycast addresses either aren’t special (that is, they are just IP addresses in someone’s prefix) or are link-specific (e.g., the subnet router anycast address, which if I understand it correctly is constructed of <local-prefix>::0).
> 
> What I am looking for is an anycast address that won’t match any local prefix, so that it filters out towards an egress router and is caught somewhere along the way, or worst case, at the egress.   I can see where this would have gone down in flames, since we don’t want anycast packets to keep going toward the backbone and create congestion, so that might explain why this hasn’t happened.   But we do have the notion of scopes, e.g. for multicast, and that would seem to apply for anycast as well.   We do allow multicast in scopes larger than the local subnet, and AFAIK this has not melted the Internet.
> 
> The actual use case I have for this is wanting to be able to have a constrained device send a unicast discovery or announcement which can be assumed to be caught and handled by infrastructure.
> 
> So, is this something that’s been talked about and abandoned as a terrible idea, or abandoned because nobody wanted to do the process to make it happen, or is it (seems unlikely) an innovation on my part?   Or is it already done and I just managed to not find the document describing it?