On Fri, 10 Aug 2001, Hayling, David wrote:
> Having just seen the latest O'Reilly book "DNS and Bind" 4th ed Apr 2001, I
> note on page 302 it says
> "Use of the old AAAA record and ip6.int is now deprecated" So who has
> changed their mind? Did the IETF originally deprecate AAAA?
>
> The O'Reilly book goes on to say on page 302 "The main reason the AAAA
> record and the ip6.int reverse-mapping scheme were replaced was because they
> made network renumbering difficult". Does the IETF have a solution to
> renumbering when the AAAA scheme is in use?
The problem is that the renumbering scenarios and mechanisms aren't exactly
well understood right now. A6 was designed with renumbering in mind, but
it doesn't sound like there's any concensus on what "rapid renumbering"
really constitites, or a firm understanding of the consequences of such a
requirement. Christian Huitema's renumbering draft presented at the IETF
is a first step.
There are questions over A6 in terms of the potential for people to
severely misconfigure things, and for A6 chains to get perhaps unmanagably
long. Of course, you can also mess up with plain A and AAAA records :-)
Some people also argue that you can use scripting tools to achieve what
A6 offers with AAAA.
The IETF meeting concensus was that AAAA should be deployed, and A6 made
experimental status. Note that's not A6 being made historic (really dead)
but strongly recommended not for production use. With many networks
rolling out initial IPv6 services, a clear message was needed, and that's
for AAAA. Whether exim chooses to support A6 is up to the developers.
The IETF also dismissed the idea of A6 but only (now) using A6 0. In
reality, A6 is as good as dead though.
We've run a number of BIND9 servers on European academic testbed IPv6 servers,
using A6/DNAME and ip6.arpa. In its simple state it works, and you can even
synthesise AAAA records from A6. But I think the IETF consensus is
right, if we want deployment to not be held up now.
I've heard talk of a major reworking of DNS altogether, but that would be
a 5-10 year process in itself (as the IPv6 introduction has proven to be).
Tim
