Re: Joint DNSEXT & NGTRANS agenda

[Robert Elz <kre@munnari.OZ.AU>]

    Date:        Sat, 04 Aug 2001 23:05:03 -0700
    From:        "D. J. Bernstein" <djb@cr.yp.to>
    Message-ID:  <E15TH27-000N5G-00@psg.com>

  | What records are you talking about?

Otha-san was talking about glue A records that accompany NS records,
which is a side avenue that this discussion was taking.

  | Typical A6 reliability problems involve servers that don't know the next
  | step in the A6 chain. How is the server going to provide records that it
  | doesn't have?

It isn't.   Obviously.   Whether or not there's a reliability problem
in this case will depend upon how reliable the other servers, that have
the rest of the data are of course.   This is no different than an MX
record that names a host served by a different server, or a similar NS
record.

And, yes, I know that you recommend to your users that they don't do that.
And that's fine.   And you can do the same with A6 as well...

  | Of course, we can eliminate the reliability problems by making sure that
  | the servers have the entire A6 chain---in which case it's far simpler to
  | use AAAA.

But if A6 is used, each site has 3 choices.   It can use A6 0, which is
identical in effect to AAAA.   Or it can use an A6 chain, where all the
RR's are available at the same server.  Or it can use an A6 chain, with
one or more of the A6 records at some other server (hence requiring more
lookups).   Each site gets to make that choice, and make it indiviually
for each v6 address to be placed in the DNS.

That is, each site gets to choose between the degree of reliability, and
the degree of flexibility that meets its needs best.

You are, of course, free to keep recommending to your users that they
only use A6 0 if you want to.   You could even (assuming you ever implement
any of this) refuse to allow your server to load any A6 record other than
A6 0.   But is there any good reason that you can give us why others
shouldn't be free to make other choices?

kre