[ppml] IPv6>>32

[Michael.Dillon at radianz.com]

> What size assignment do you advocate for a produce crate?  Please
> extrapolate the lifetime of IPv6 if consumables get subnets.

I would use a /64 per crate. Each crate will have a network of 
somewhere between 6 and 12 sensors monitoring air temperature, 
humidity, condensation/juice on the crate floor, concentration
of ethylene gas, and a 3-axis accelerometer. There will likely
also be an LCD panel as well to identify the contents and
physical destination.

> What kind of addressing policy and routing system would you propose
> to scale to that kind of network?

Much the same as the existing IPv6 policy except that I
think we could have a cap on the HD ratio to limit the
size of a single allocation. If people actually do outgrow
an allocation then they can come back later and get more
at a time when we have more data and experience. As for
routing, I think you know about my support for a geographical
addressing hierarchy outside of 2000::/3. This would be
based on a detailed plan for address deployment covering 
the entire globe in a way that could be fully agregated
along existing fiber pathways. That means it ignores national
borders and sees the world as a collection of cities/towns
connected by greater/lesser quantities of fiber. And this
plan would also only use a fraction of the IPv6 space no
larger than twice the size of 2000::/3. The intention is
that the current addressing plan and the geographic plan
should compete in the marketplace and in a decade or so,
when we know more about what really works and how things
are developping, we could revise or replace both plans. 
In the interim, the geographical addressing plan minimizes
consumption of slots in the global routing table.

The most important aspects of this are that there are
two distinctly different variations of addressing plan 
that compete and the IPv6 space has enough reserve to
completely replace both plans and still have space left
in reserve. This is the only plan that can scale. People
who have direct experience with scaling either networks
or businesses or chemical plants know that unforeseen
factors cause you to change your assumptions and change
your plans. We need to expect the unexpected and plan
to be surprised and plan enough reserves to adapt to those
surprises.

> > I think we are allocating less than in the past. In IPv4 we give 
> > a new ISP 20 bits of address space. In IPv6 we give him 32 
> > bits in his prefix. Therefore the IPv6 ISP is getting a much 
> > smaller fraction of the total address space than the IPv4 
> > ISP. These people who talk about waste simply do not 
> > understand IPv6 fundamentals. Either that, or their 
> > definition of "waste" doesn't match what I read in the dictionary.
> 
> A smaller fraction for ISPs, but as you point out, there are many
> different kinds of entities that could get assignments.  It seems to
> me that most arguments assume a higher rate of growth than the 
> current curve.

Nevertheless, today in IPv4 we give a /32 to a single interface on
a single device, sometimes mistakenly referred to as a "host". In IPv6
we give that same /32 to an organization running a large network
with at least two levels of hierarchy, namely the organization's own
network and the networks of the end-sites. In many cases, these
end-sites also have 2 levels of hierarchy, but not in all cases.

This is already very much more scalable than IPv4.

--Michael Dillon