I've been thinking about a not yet mentioned (AFAIK) solution for a while but dismissed it as too ambitious. Yet, the longer we keep going without getting there, the more reasonable it seems to me. I suppose the respons will be "too much work, too late" but it could be worth considering at least.
We could consider wiping the slate clean and design a new protocol. (Bear with me just a little while longer.) Let's call it e g INS (for Internet Name System or International Names System) to distinguish it from DNS and ask IANA to assign, say, port 40 for it's use. We would not only be able to introduce non-ASCII domain names using UTF-8 in a nice, clean fashion and in accordance with RFC 2277. We could double (or triple) the maximal allowed length for domain names and have four times the number of flags. We could introduce a version number in the protocol to make future changes easier and use experience with EDNS to make it easier to expand in the future if needed. We could replace SRV records (RFC 2782) with a cleaner and better solution, introducing QSERV and QPROTO fields in the queries. And so on. All in all, we could clean up and improve the current DNS standard significantly. It seems we have a golden opportunity here. Nobody will use a new protocol to get SRV records or longer domain names, but perhaps to get international domain names. Does something like this really have to take that long time to get consensus on, especially compared to the time already used without reaching consensus so far? If we stick to basic DNS principles and simply change the format of RRs, I think we already have a pretty good idea (gained from the past years) about what needs to be done; as long as we are careful and don't try to build a completely new system but focus on cleaning up the old one. In contrast to IPv6, this could very well be used in parallell with the old one and handled by the same software. Magnus
