I suspect we have 3 alternatives:
b) IPv6 with aggressive prefix length filters and highly indeterminate
reachability for longer prefix PI address holders.
c) IPngng (maybe IPv6 with some sort of locator/id split hacked onto it)
The default will be (a).
it's not at all clear that these are mutually exclusive.
IPv4+NAT is here and it's not going away anytime soon. Even if an ideal
solution using IPv6 were to surface tomorrow, we'd still be dealing
with IPv4+NAT for 10 years or so. The most popular applications that
exist today will be the last ones to move to IPv6 (or to stop supporting
IPv4) because those are the ones that have the most investment in IPv4
and NAT workarounds. To put it another way, IPv4 is part of the
critical infrastructure for those applications. They might start
supporting IPv6 in addition to IPv4, but IPv4 support will be the
necessary condition for interoperability of those applications until
IPv6 is as ubiquitous as IPv4. Meanwhile there will be a need to host
applications on IPv6-only networks that still have access to the global
IPv4 infrastructure - and the solution to this probably ends up looking
a bit like a NAT (though one in which apps are explicitly aware of and
control the bindings rather than something like NAT-PT that tries to
fool apps into thinking it isn't there).
IPv6 with PI addresses seems likely - the only question is how big you
have to be to get a PI address prefix. In the near term, scalability is
not an issue, but if IPv6 is at all successful the growth in routing
table sizes, updates, etc. could be quite steep. Prefix length filters
might not be the mechanism used to decide which prefixes get routed and
which ones don't, but there will be some mechanism for this.
Lots of versions of LOC+ID split have been talked about over the years.
A lot of people believe in the concept, but the devil is in the
details. Still, my impression is that newer proposals in this area are
more realistic than those I saw a few years ago, both in actually being
implementable and in accommodating the diverse set of interests that is
the Internet. I think it will still take a few years for a
standards-quality solution to emerge. Meanwhile, we'll be using some
mixture of the above. We might or might not get a good LOC+ID solution
in place before the routing scalability limitations of PI addresses
result in another crisis.
But I really don't think we have the luxury of choosing one of these
over the other. We need to work on all of these, and more. We need to
think of them as complementary approaches rather than competing ones,
even while we recognize that some of them have much better long-term
viability than others.
Ietf mailing list