Re: Generic anycast addresses...

[Brian E Carpenter <[e-mail]>]

On 31-May-19 05:02, Ted Lemon wrote:
> Thanks, Sander (and everyone else!).   I’m somewhat inclined to just write a document that specifies a prefix to use for site-scoped anycast addresses.   This would give us a document to which to propose changes.   I think that the discussion thus far has been productive.   The point of this is not to override the work that Mark is doing.   It’s just to do a bite-sized chunk that has a clear use case.   I’m not convinced the WG will adopt it, but unless somebody has something new to add, I think what needs to be said has been said.
> 
> I think the two spaces out of which this could come would be fec0::/10 (site-local) and fc00://8 (the undefined block of the ULA prefix, fc00/7).   I don’t have a strong opinion on this; the arguments against using site-local that I can think of are:
> 
>   * People might still be using it (that’s okay, this is a new application)
>   * GIven that it’s been deprecated for a long time, nobody is using it, so we should just mark it available for allocation, and then this isn’t the right use for it (I think this is a good argument against, in theory, but don’t have any personal opinion on whether it is correct).
>   * Site-local addresses would in fact be routed onto the backbone (I think this is a bad argument against, but maybe I’m wrong).
> 
> 
> The argument against fc00::/8 that I can think of is simply that we should leave it available for future use as originally intended.   I think you can’t argue that it shouldn’t be used because it might be routed on the backbone, because if that’s true, it’s a violation of current expected practice and should be fixed.

I just looked at RFC3879 for the first time in many years.
Especially https://tools.ietf.org/html/rfc3879#section-2 "Adverse Effects of Site Local Addresses". I think the arguments there are still interesting.

Ted, please do think about the main point: "the fuzzy nature of the site concept". (And, shameless plug, see https://tools.ietf.org/html/draft-carpenter-limited-domains). Limited scope anycast needs a non-fuzzy scope boundary.

   Brian

> 
> Of course, the other argument against this is simply that we shouldn’t do it, but I haven’t actually heard anyone say that, so that’s encouraging.
> 
> Thanks very much for the discussion!
> 
> 
> --------------------------------------------------------------------
> IETF IPv6 working group mailing list
> [e-mail]
> Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
> --------------------------------------------------------------------
>