[arin-ppml] Encouraging IPv6 Transition (was: Clarify /29 assignment identification requirement)

[Owen DeLong]

On May 15, 2012, at 11:56 AM, William Herrin wrote:

> On 5/15/12, Chris Grundemann <cgrundemann at gmail.com> wrote:
>> On Tue, May 15, 2012 at 12:00 PM, Michael Richardson <mcr at sandelman.ca>
>> wrote:
>>> And I'll say it again: enterprises and equipment vendors need
>>> Non-Connected Network space, and they need it at the same cost as
>>> RFC1918 address space.
>> 
>> It's RFC 4193 space (ULA) for IPv6:
>> https://tools.ietf.org/html/rfc4193. Or am I missing something?
> 
> Hi Chris,
> 
> The math for statistical uniqueness in ULA, while internally correct,
> is based on some suspect assumptions. If you replace them with
> worst-case assumptions, the probability of collision when
> interconnecting two large organizations increases to something on the
> order of 1 in 1000. Maybe higher if you consider human factors as
> well.

So what... He said he wanted equivalent functionality to RFC-1918 where
your risk of collision is more like 1 in 3 at best and usually 1 in 1 in my
experience.

> ULA Central or a similar RIR-managed ULA space might provide a better
> guarantee of uniqueness and would, for a certainty, make it easier to
> figure out who's "in the wrong" and must renumber when a collision
> occurs. It would also offer a user better control over what happens
> during data leak scenarios, e.g. RDNS requests which incorrectly make
> it to the Internet DNS servers.

It would also become GUA in short order.

> On the flip side, ULA is at it's very worst still far better than RFC1918.

Exactly. It's still a bad idea, but, it's a less bad idea than RFC-1918 was
and not nearly as misguided as ULA-global.

Owen