On Wed, Nov 28, 2001 at 03:35:21PM -0500, Keith Moore wrote:
The situation today with NAT is that hosts in separate realms can only
communicate in 99% of the desired applications,
to the extent this is true, it's only because the only applications
that people become aware of, are those that can run over NAT. many
more useful applications exist, but since they can survive only in
less restricted environments, they aren't as well-known.
I agree entirely; I will just note, <sarcasm>curiously</sarcasm>, that the
"explosive growth" of the Internet coincides with the widespread desire for
access to a *few* "client-server" type applications. It's not clear that
growth in those applications cannot be satisfied by NATized environments. That
said, the network cannot be designed for just that paradigm of application.
I'm hoping there will be a different class of "killer application" that will
attract many more people to the Internet and more than likely, I suspect it
will _not_ be amenable to a NATized environment.
if you're willing to constrain those billions of addresses to use
a single path to the net (as NAT does) then the existing routing
system does just fine.
We could also hope that the next set of "killer applications" be more
adaptable and carry sessions over multiple paths with multiple addresses at
the endpoints as end-host multihoming is likely to be more scaleable than the
present regime. That would mean not having to give up multihoming for future
applications and having to build shims for current ones if no "magic bullet"
routing system is found for IPv6 that allows multihoming as familiar as is
common under IPv4.
Anyone have an idea for robust source-address selection? anyone?
Adi