[dns-operations] TLD(s) for private use

Suzanne Woolf suzworldwide at gmail.com
Thu Sep 7 20:47:32 UTC 2017

On Sep 7, 2017, at 6:27 AM, James Stevens <James.Stevens at jrcs.co.uk> wrote:

> And https://www.ietf.org/id/draft-ietf-dnsop-sutld-ps-08.txt puts all the meat on those bones, I look forward to the RFC, please CC this thread, or me personally, if you remember.

It'll appear when the RF Editor is done with it, sometimes a little hard to predict but several of the folks who've been participating in this thread took part in the DNSOP discussion on the document, so I'm sure one or more of them will forward the pointer when it's available.

>> Some issues are just like the challenges with RFC 1918 IP addresses;
>> some are different, at least in part because people care both about
>> having domain names that are unique within a particular scope, and
>> about which names those are-- short strings, or "words," or IDNs. (IP
>> addresses are fixed length and people are usually, although not
>> always, indifferent to which numbers they get, aside from concerns of
>> routability and aggregatability.)
>> ... [snip] ....
>> Part of the challenge is that domain names aren't used only in DNS.
>> Another part is that the DNS protocol assumes a global or default
>> context for resolution, so it's hard to indicate what "local" means.
> These two points, when taken together hit a number of crucial issues at play here.
> I think its also very interesting to bring IDN into the mix, as I'd not thought of that.

It's an extension of the observation that domain names are used everywhere there's an Internet, and in most of those places people think of domain names as "words" and value them in some of the same ways they value natural language objects. Much of the pain people get into around domain names results from those expectations and the gap between them and reality. (The "problem" that DNS is exact-match and doesn't support fuzzy matching is somewhat related: people would like that, or think they would, but it's not a property of the DNS and is probably met better somewhere else.) 

A couple of comments upthread also point indirectly to the additional complexity that's specific to single-label names, which is that the root zone is a globally shared resource so there's a fairly large, complex, bureaucratic set of processes involved in deciding what TLDs will be created--  or even what TLDs *won't* be, which is why it's hard to be sure a given string will never ever be delegated in the production DNS as a TLD. 

It's also got to do with the scalability of any given approach: if you think it's important to have wide-scale conventions that meet your expectations of domain name uses and behavior, it's pretty important to assume others will feel the same way about their expectations, including that they be able to express their company's or pet's name in an IDN. 

If you want a standard, it won't work for everyone all the time, but it should probably work for as many cases as possible.


More information about the dns-operations mailing list