Re: (ngtrans) Re: ipv6-smtp-requirement comments?

[Erik Nordmark <[e-mail]>]

> RFC 2893 explains how hosts without IPv4 addresses can communicate with
> hosts without IPv6 addresses. RFC 2893 observes that compatibility with
> the existing IPv4 universe is essential for the success of IPv6.

Dan,

What you are describing is impossible in the general case.
In order for an IPv4-only node to be able to communicate with some
other node at the IP layer that other node needs to have an IPv4
address - the IPv4-only node needs to put something into
the destination field in the IPv4 header. 

So RFC 2893 doesn't claim to solve this impossible problem.

There are outlines of possible approaches that were at least outlined in
the now expired  draft-ietf-ngtrans-dstmext1-aiih-00.txt
and draft-ietf-ngtrans-siit-dstm-00.txt.
The idea in those drafts is just to not permanently allocate an
IPv4 address to identify the IPv6-only nodes that are communicating
with IPv4-only nodes, but instead do this only when such communication
is occurring or believed to be about to occur.
For communication from the IPv4-only node this can be accomplished by
deferring the allocation of that IPv4 address until there is
DNS lookup for an A record for the IPv6-only node.

Of course, this approach is subject to DoS attacks by just doing DNS lookups
of A records for IPv6-only nodes, so it isn't clear to me that it
will be that useful.

The alternative is to move the problem up the protocol stack
e.g. by having application layer gateways (http proxies, mail relays, etc)
on dual IPv4/IPv6 boxes and somehow configure things so that the
communication between IPv4-only and IPv6-only nodes go through those
boxes.

   Erik