[arin-ppml] An article of interest to the community....

[Paul Vixie]

On Tue, 30 Aug 2011 16:06:11 -0400
"Mike Burns" <mike at nationwideinc.com> wrote:

> ...
> And this is in answer to the question posed by Mr. Vixie, which
> postulated a no-option endpoint at IPv6.

looking at that again:

On Tue, 30 Aug 2011 12:01:14 -0400
"Mike Burns" <mike at nationwideinc.com> wrote:

> ...
> As to your issue about what happens after IPv4 is at its best and
> highest use, I have given this some thought over the years.
> Have you considered that a robust trading market and CGN might buy us
> enough time to come up with some kind of backward compatible
> successor protocol to IPv4?

yes, i considered (and rejected) this approach back in the late 1990's.

> Proposals for these kind of solutions have been made in IETF but are 
> routinely ignored.

there are several reasons why this kind of solution is routinely
ignored when it comes up in the IETF or elsewhere.  first, because the
internet's initial success depended on a stateless core, and most of us
assume that to make traffic capacity scale with transistor density we
will have to continue keeping state out of the core.  the internet
core is packet based even if the web we've built atop the
internet appears to be virtual circuit based (tcp/80).  second, because
the transition from ipv4 to ipv6 is already very difficult and most of
us know that leaving basic design questions open would make the
transition even more difficult.  (we now know better ways to do secure
dns than the current dnssec design; we now know better ways to do
packet switched networking than the current ipv6 design; but we have to
stop designing and stop building at some point, not optimize forever
and get nothing.)

> To my mind there is room in the IPv4 header to expand the effective
> address space.
> GGN does this effectively and largely transparently today using the 
> additional room in the  port number part of the header.
> Every TCP connection uses IP+port, and we have 65,000 ports per IPv4 
> address.

if the internet were not older and larger than the web, then your
web-centric mental image as outlined above might be a realistic way
forward.  notwithstanding the fact that a lot of us build a lot of
services on top of tcp/80 since it's a fairly reliable serial channel
that works even in hotel rooms, the internet also has to support packet
based services that do not map well to virtual circuit style mechanisms
like tcp.

> The dual-stack transition is a failure, it is time to recognize that
> the lack of an economic motive for IPv6 created a situation where
> dual-stack was doomed.

disagreed.

> And we have gotten pretty darn good at traversing NAT already,
> despite the doomsayers.

also disagreed.  what "we've" done is hold back some kinds of
applications altogether because of NAT, and increase both the
development and operating costs of other applications because of NAT.
i am treating NAT as a transitory economy-wide externalized cost, not
as an inevitable part of the future of all IP networking.

> It does not further your position to openly wonder whether those who 
> "complain" are "hoarders and speculators".

noted.  however, the thing you're complaining about here isn't the
position i was describing.  if you only wanted to sell "web access"
rather than "internet access" then you can live with NAT and you'd have
no oar in the water wrt IPv6.
-- 
Paul Vixie