Re: Generic anycast addresses...

[Ted Lemon <[e-mail]>]

On May 31, 2019, at 2:11 AM, Mikael Abrahamsson <[e-mail]> wrote:
> https://tools.ietf.org/html/rfc6052#section-2.1 <https://tools.ietf.org/html/rfc6052#section-2.1> , the 64:ff9b::/96 prefix for NAT64, out of global scope. However, you typically need the traffic to only hit one of these (because of state), so I guess they will not be BGP anycasted.

That’s a similar idea, but not anycast.   There you’re hoping the packets reach an edge router that does NAT64.   So a lot of the work in the infrastructure is the same—if you have the server at the edge and only one egress, you don’t really have to do anything except make sure that the edge router forwards the packet to the right host.  If you want something fancier, you set up the right route—in the case of 64:ff9b, if you want your NAT to be somewhere other than the edge, you just arrange your internal routing so that the packets get to the network address translator.   You could do the same thing for these anycasts.

One disadvantage of allocating well-known anycast addresses out of the same address is that if your anycast servers are actually in different places, you need to have host routes in your routing fabric, or use tunnels, so that’s a downside that doesn’t exist for the NAT64 WKP.