Re: [ga] Re: IPv6 and root servers

[Thomas Narten <narten@xxxxxxxxxx>]

Karl Auerbach <karl@xxxxxxxxxxxx> writes:

> Do you happen to know if there are any measurements of the IPv4 and IPv6 
> root query rates?

Since there are no IPv6 root servers, it's kind of hard to measure the
query rate. Exactly _what_ IPv6 addresses would resolvers be sending
queries to?

> I guess I'm also looking for two distinct measurements, which would be 
> queries of any kind carried *in* IPv4 vs IPv6 and, separately, queries 
> *for* IPv4 or IPv6 records (or either).

I would not expect the types of queries to vary from IPv4 to IPv6
transport. The relevant standards are all transport agnostic. I.e.,
you query for A and AAAA records separately, and which transport is
available does not factor into the type of queries that are generated.

> Another interesting bit of data would be the percentage of queries that 
> had the truncated bit set and how many TCP based queries occurred.

Very few to none, would be the expectation. Indeed, I think DNS folk
have a general concern that should responses get truncated, the fall
back to TCP would not work in all cases, because the code to do so has
simply not been required in practice, and thus is not well-tested in
practice, which almost certainly means issues will crop up should it
become necessary to use TCP. E.g., (and I'm just speculating) there
are NAT boxes that think they know all about DNS queries and how to
muck with them. It is unclear that they will do the right thing in all
cases with DNS over TCP.

> The reason I'm interested in that is I have concern that as IDN
> names (which tend to require more bytes to express) are increasingly
> used that those roots (and TLD servers) that don't do EDNS (are they
> any?) will send back truncated responses that may cause TCP based
> follow-up queries.  I'm concerned that there might be - and I'm
> merely speculating that there might be - some kind of inflection
> point in which we might find that suddenly the load on DNS servers
> increases because of a UDP to TCP transition resulting from IDN
> names.

I think this concern is shared by others. It is one reason why folk
are being extremely careful about moving the system into a regime
where TCP starts getting used.

Thomas