[ppml] IPv6 initial allocation policy

[Randy Bush]

> 1) The number and density of hosts in IPv6 networks is & will be
>    irrelevant.

once upon a time, we thought that of v4.  luckily, history never
repeats.

> 2) We need to work within the constraints of the existing BGP
>    protocol for the foreseeable future.

thanks for that.

> 3) As long as IPv4 is run in parallel, the number of subnets will
>    be the same because it would be too hard to explain to ops how it
>    works otherwise.

this seems to assume that the v6 net is essentially congruent with
the v4 network.  perhaps this assumption is worthy of exploration.

> 4) If a subsequent allocation needs to be made, it should
>    aggregate the current one, not just be adjacent (6.5.8.3 needs
>    work)

like the thousands of de-aggregated adjascent /24s are aggregated
today?  not.

> 5) The need for PI space has absolutely nothing to do with the
>    size of the network.

this is a political, not an engineering statement.  while politics
should not be ignored, social theories which are a radical
departure, such as every home should be pi, deserve suspicion.

> 6) The only reason for having any measure is to preclude the
>    masses from taking global routing slots; though specifically RIR
>    policies 'say nothing about routability of the assignments'.

the reasons for prudence have nothing to do with suppressing the
proletariat.  they are based in a history of problems with address
space turning out not to be infinite and routing churn turning out
to be a serious problem.

> 7) There really is no single global DFZ even today, so approaches
>    that allow pockets of aggregation 'as needed' will not break
>    anything.

can you describe examples of these "pockets of aggregation" as
deployed today?

> 8) The number of PI entries and the capabilities of routers will
>    evolve over time, so whatever approach is taken now it should be
>    clearly identifiable and allow for future aggregation of early
>    assignments if/when/where the need arises.

for us to bet on this, the protocols and engineering also "should
be clearly identifiable and allow for future aggregation of early
assignments."  are they?

> In the mean time we could discuss the relative importance of
> putting something in place before it becomes an issue for the ITU
> to bolster their drive to take over global IPv6 assignments.

or the chinese.  muslims.  or the christian right.  or the
boogeyman.

can we focus on the engineering discussion?

randy