NATs and addressing

Thu Sep 11 22:22:34 UTC 1997

Ok so this turns NANOG into Big-Internet... mutter... but
it's more interesting than television, and people here
really ought to be thinking about things like this anyway.

Oh yeah, I just realized I can add an
ob. ops. postscript.  So there.  Nyah.

Michael Dillon <michael at priori.net> writes:

> If you use NAT and renumbering technologies then you
> don't give a darn what your IP address is or who gave it
> to you as long as it is globally routable.

No, all you care about is that two endpoints can talk to
each other.

You emphatically must NOT care what your IP address is, as
you say, but this also means that you must not care about
the scope of the address itself, nor about the scope of
its routability.

In any piece of the evolving Internet, a datagram may
undergo transformation from one address to another.  In
some cases it may undergo transformation into a prefix
which is expressly not globally routable (e.g., an RFC
1918 address), yet which is routable within the scope of
the address's visibilty.

That is, between any pair or cluster of NATs (for
instance, at the edges of a network) you require
routability of all addresses valid within the scope
delineated by those NATs.  Outside that scope, the
addresses (and indeed the protocol(s) under the end-to-end
protocol) may be undefined.

> Routing problems involve how to design, manage and operate this internal
> traffic distribution hierarchy and are essentially engineering problems,
> not policy problems or social problems. I think that the desire for
> portable address space is not a routing problem.

PI space is already being remapped by NATs around the
world; the next steps in the evolution of the Internet
seek to push that remapping closer to the edge of the PI
space, so that the PI addresses ultimately are visible
only within the administrative scope of the user of the PI
addresses.

This done, the whole PI vs PA address argument will simply
cease to be relevant, and the principal flaw of the
post-CIDR IPv4 addressing scheme will be no more.

Note that scoping address meaning within a subset of the
global topology also increases the address lifetime of
IPv4 itself substantially, even ignoring such things as
overloading single addresses and the like.

The one piece that is unfortunately missing is more an API
issue than anything else, and that is that getservbyname(3)
should take an FQDN as a third argument and should find
its answer in the DNS rather than in /etc/services (hi Paul).

	Sean.

P.S.: The neat thing about this is that suddenly no
	operator really would need to worry much about ARIN and
	addressing politics, since addresses would only be
	valid within paritcular scopes, which could be the
	individual provider or a collection of providers,
	with or without downstream customers.

	(I hope there is no operator who thinks that
	killing off silly addressing politics and the
	endless arguments about filtering and so forth would be a
	Bad Thing.)