John's email convinces me that the whole DNS issue is a red herring.
First off, nobody is ever going to set up IPv6 only email, not while there is
any appreciable base of IPv4. They might add IPv6 support in addition.
Fortunately this is really easy to support. Let us imagine that we have two
separate servers for the IPv4 and IPv6 support. The sysop writes an MX record
example.com MX 1 1 1 smtp1.example.com
smtp1.example.com A 10.1.1.1
smtp1.example.com AAAA ......
A legacy IPv4 only SMTP server automatically chooses the A record and hits the
IPv4 server. The IPv6 capable server chooses the AAAA record.
It seems to me that this is going to be the story for every useful application.
We do not have to make a fetish of doing everything with IPv6 alone. That is
not the world we are moving to. In the real world of the future my home machine
is going to continue to outsource SMTP mail submission and DNS query resolution
to external servers. It is quite possible (certain even) that at some time in
the future the pool of IPv4 addresses will dwindle to the point that my house
does not have its own unique IPv4 address and so the communication between my
home machines and the outsourced SMTP/DNS servers takes place over IPv6.
We have 4 billion IPv4 addresses. We are most unlikely to ever get to the stage
where we are so short of them that an ISP cannot find enough IPv4 addresses for
their servers and other proxies.
So what if you can pull up the .com domain via IPv6? The DNS server still has
to be IPv4 capable or the query will quickly fail at microsoft.com, google.com
From: John C Klensin [mailto:john-ietf(_at_)jck(_dot_)com]
Sent: Fri 04/01/2008 3:46 PM
To: Bill Manning
Cc: Barbara Roseman; ietf(_at_)ietf(_dot_)org
Subject: Re: AAAA records to be added for root servers
--On Friday, 04 January, 2008 12:01 -0800 Bill Manning
Again, there is no specific root server issue in any of this,
but it concerns me that none of the relevant committees or
studies appear to have considered the possible applications
implications of the change.
actually, there was on at least two occasions the explicit
scoping of the consideration to the DNS-specific issues.
interactions with other applications was ruled out of scope
regarding the issues of adding what is effectively a new
address family to the DNS. The general answer when needing
to communicate between similar applications that run on
different address families has traditionally been the
application layer gateway (ALG) ...
Then, IMO, whoever ruled it out of scope should look for a job
describing large animals by touch and while blindfolded. If
said large animal responds to inappropriate poking by stomping
on the individual we can make a nomination for the Darwin Awards.
Seriously, a competent SMTP server running in a competent
dual-stack environment should be nearly agnostic about whether
connections are occurring via IPv4 or IPv6. There are several
server-environment pairs that are that competent. Most major
contemporary SMTP servers have their origins in multiple
transport environments (not just different address families) and
RFC 821 was explicitly written to recognize and accommodate
multiple transport environments. If we are now making
decisions about IPv6 deployment that effectively force the use
of an ALG, rather than servers that can handle both
environments, we have taken the fine art of shooting ourselves
in the foot to new heights.
I also note that, even if one ignores "other applications", the
DNS has already recognized email as a special case, with its
own, applications-protocol-specific RR type and rules --both
rules within the application protocol and rules about additional
information that is returned with MX queries. If the MX
resolution doesn't work smoothly for IPv6, then the DNS isn't
IPv6-ready no matter how many AAAA records are defined and
spread around. Even if one thinks an ALG is needed, one has to
find the ALG host. I hope the additional information rules have
been adjusted if needed to encourage return of relevant AAAA
records if they exist: it hadn't even occurred to me to check
until now, but your comment about ALGs forces me to wonder
whether DNS changes for IPv6 have ignored the fact that there is
an Internet out there on which people run applications.
Ietf mailing list
Ietf mailing list