wow, that was a dense mail, at least it's formatting :-) Just wanted to add, since there was a question about promoting the deployment of EDNS0, that some of us do indeed try to work on that line because it would make some matters simpler (eg. adding v6 glue without fear, having more root servers if needed, etc). DNSSEC requires the use of EDNS0 so maybe some people will migrate if DNSSEC takes off. Currently, some of us monitor the amount of queries using ENDS0 that reach the servers we operate. At least on F root this figure is more or less stable between 30-40% One thing to keep in mind is that there are products shipping today which are based on some version of BIND4, for instance, so they support DNS as it was known in the BIND 4 days. Although we have approached some of the manufacturers, they can't just "transition" in a short time. In the meantime life goes on and there is a need to cope with a growing Internet. Hope this helps clarify things a bit Joao On 15 Jun, 2004, at 19:18, Jim Reid wrote:

Not having seen the whole discussion thread, it's a bit hard to make sense of what's been said. However it appears to be that someone is saying that the need for anycasting (and more specifically special address allocations for anycast name servers) can be discounted if ENDS0 was more widely used. This is a flawed argument IMO. As Peter has said, resolvers that aren't EDNS0-aware will still be pounding on the parent zone's name servers. [They're also much more likely to be the resolvers that are misconfigured to go to the root to reverse lookup RFC1918 addresses, don't implement negative caching and so on....] Simply adding more NS records and glue in the parent zone's delegation for a child zone is no help. For one thing, most name server implementations have a limit on the number of name servers they can handle for a delegation. Adding extra NS records is even less help when those servers have IPv6 addresses => yet bigger DNS payloads. In fact adding extra servers and/or IPv6 addresses may be worse because there's an increased likelihood of truncated responses getting sent to these non-EDNS0 resolvers, resulting in retried queries over TCP. Nasty.

Aside from these DNS protocol issues, there are plenty of other good reasons for deploying anycasting for important DNS infrastructure. That's why lots of the root and TLD name server operators are doing this already. Ironically, these include the NCC's root name server. Anycasting provides increased robustness, extra redundancy, improved performance, better scalability, extra capacity/throughput, defence in depth from DDoS attacks, etc, etc. Anycasting isn't going to go away even if all the world's DNS software implemented EDNS0. Anycasting is a fact of life. And it will become more prominent in future.

So if the address policy WG is reluctant to endorse special address allocations for DNS anycasting, I'd ask them to reconsider. If it helps, we could ask the DNS WG to discuss the issue and perhaps make a recommendation to the address policy WG.