[address-policy-wg] IPv6 PI request is turned down for my multihomed hosting facility - Why?

Hi,

On Wed, Mar 30, 2011 at 7:14 AM, Gert Doering <gert at space.net> wrote:
> On Wed, Mar 30, 2011 at 11:54:55AM +0200, Jan Tuomi wrote:
>> Setting up an SSL-webhost is also sub-allocating?
>
> This is very grey area.  Technically, it's "giving an address to a 3rd
> party", but personally I'd see this is as "it's still the same machine
> under *your* control".

Right. I want to take this further by revisiting some core networking
principles with a touch of v6:
Control of usage of addresses, even in times of /64 v6 SLAAC-enabled
LANs, remains with the party that switches and routes them, normally
but not necessarily, to and on the Internet. This is the reality.
Consider this for a while.

APWG is fooling itself in thinking that it can control what addresses
anyone configures on a host. Influence (in only a small way), yes.
Control, no.  The core registry function of RIPE *is* to facilitate
Internetworking by avoiding addressing collisions.
Put different, RIPE concerns itself primarily with assisting in
organizing addressing, allowing networks to communicate, *not*
specific hosts.  The networks concern themselves with making hosts
communicate. A host can have any address configured, free of
consequence, until it is connected to a network.

Much of the pain from APWGs policies comes, IMO, from the distance of
the disconnect a policy has with reality.
  The further the disconnect, the greater the resulting pain and lies.
 Close the gap, reduce the pain and lies. PIv6 has a very big
disconnect with reality right now IMO, which lately has become very
evident.

Tieing the "user" of addresses (in RIR sense) closer to the party that
routes and switches them(*), ie the network, and away from the ones
who configures them on a host stack, is 100% in line with RIPE's role
as a registry. In IPv6 with SLAAC, this is particularly true.

* For administrative reasons, it becomes at some point very burdensome
to track usage (in RIR sense) at the RIR level of individual end-users
(ie, home users) which is why a lower threshold of "allocations" in
IPv4 of /30 was established long time ago. I believe the equivalent
value in IPv6 is /48, which in itself probably is subject to
considerations w.r.t BCP157 now.
Conceptually, each organizational level in the addressing hierarchy
should (for abuse- and law-enforcement purposes, I suppose..) be able
to provide a pointer, a next-hop to the next entity in charge of
something.  In theory, RIPE's only pointer to resources delegated to a
LIR could be to the LIR itself (who of course is free to keep its
record in a public database such as RIPE's).

> But we already know that datacenter has all the range from "very obviously
> *not* sub-allocation" to "very obviously this *is* sub-allocation", and
> as such, distinction between "what is OK" and "what is not" is tricky at
> best.

So let's substitute it for something that is clear, logical and
consequential. :)

Regards,
Martin