On Tue, Aug 02, 2005 at 06:30:07PM +0900, JINMEI Tatuya / [EMAIL PROTECTED]@C#:H wrote: [...] > Even if I were to choose to use IPv6 for some reason, it seems I'd be > just happy with ULAs for such purposes, and then the (revised) default > address selection algorithm seems to suffice (and so we don't need the > automatic configuration mechanism). I didn't understand why the > default algorithm doesn't work here after re-reading the past message > you attached and re-reading Stig's message. Perhaps I simply > misunderstood the point - so could you provide a concrete example > explaining why the default rule doesn't suffice there?
My point is perhaps not that important, but here's what I was thinking of. Assume you are two sites are connecting to the global internet, both using ULAs with a ULA peering between them. Now, if a host is one of these sites do a DNS lookup for a host in the other site, and get both a ULA address and a "normal" global address, you will probably want to prefer both source and destination ULA addresses. However, if a host in one of these sites do a DNS lookup for some random host on the global internet and for some reason gets both ULA and "normal" global address, it should prefer the "normal" one. It might be discussed how important the latter is. The connection attempt would hopefully fail quickly. Also sites are not supposed to leak ULAs (should use split-DNS probably), this is almost bound to happen. Perhaps it's good not to require split-DNS? I have another point regarding ULAs and multicast. As discussed earlier, there is a problem with longest common prefix match prefering ULA source address for multicast. We've discussed a change saying that the "normal" global address should be prefered for global multicast. I believe for many sites using ULAs it would be natural to use ULA source address for scopes <= 5, but there are other sites where it would be natural for <= 8. With ULA peerings between different organizations and sites, it is possible that some may want ULA source for say scope <= A, while "normal" global addresses for larger scopes. What I'm trying to say is that the part of the network that uses ULA and have ULA peerings may not map to the same multicast scope in all deployments. Stig -------------------------------------------------------------------- IETF IPv6 working group mailing list ipv6@ietf.org Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------