[ppml] [address-policy-wg] Those pesky ULAs again

[Stephen Sprunk]

Thus spake "Thomas Narten" <narten at us.ibm.com>
>> sounds like a great idea for all of ipv6 allocation.  what is the
>> difference ula or pi/pa?
>
> Here's my take.
>
> ULAs are not intended to be publically routed by ISPs. While
> some may attempt to get ISPs to route them, ISPs will have
> clear documentation saying they are not intended to be used
> that way, and they are free to filter them. And in fact they
> SHOULD be filtered. (I'd say MUST, but since that is not
> enforceable...)

ISPs SHOULD filter RFC 1918 space, but many don't.  If it weren't for the 
collision issue, you'd be able to get to a heck of a lot of "private" 
networks.

> We have ULAs already. What is missing is centralized ULAs.
> I've had enough conversations with people that want to use ULAs
> - but simply aren't satisfied with probalistic uniqueness. They
> want something more meaningful, like a signed contract that
> that can pay some fee for and get some assurance that no one
> else is going to get that address. This sort of thing makes
> business people happy. People do worry about collisions and
> the impact that would have.

The RIRs are happy to sign a contract and take a fee to issue unique address 
space -- it's called PI.  All you're going to do is make the RIRs allocate 
single-sized blocks out of fc00::/8 instead of 2000::/3, creating a v6 swamp 
that will inherit all of the problems of the v4 swamp.  There is absolutely 
nothing else different about the addresses you propose handing out, except 
you _hope_ that ISPs won't route them (much) and you _hole_ that the RIRs 
will set the requirements and fees lower.  The policy process can't 
guarantee the former and may not deliver the latter; OTOH, it could just as 
easily deliver the latter for PI if desired.

> ULAs are intended to be much more easy to obtain than PI
> space, because PI space is intended to be publically routed.

If PI is tough to get from your RIR, change the policy.  ARIN makes getting 
PIv6 space trivial for a minimum-sized request (the same size you'd get with 
ULA) and even has explicit policy stating that requests for private use are 
valid (though, for v4, RFC1918 is encouraged).

We have no shortage of address space in v6; policies that discourage PI are 
_only_ justified to the extent they are needed to keep routing table growth 
under control.  In the case of private use, that limitation doesn't apply 
since anyone who _really_ intends their block to be for private use won't 
consume a global routing slot.

> PI space, on the other hand, is not useful if it is not publically
> routed (generally speaking). Poeple obtaining PI space are
> very much assuming it will be publically routed.

PI space is useful for anything that needs a globally-unique address that 
isn't tied to a provider.  Whether you advertize it (or portions of it) 
publicly is up to you.  The same would be true for ULA-C, except there's an 
upproven theory that many people won't accept the route.

> ULA space is useful even if not publically routed (and is
> intended for uses that do not require public routability). E.g., it
> can be used to number infrastructure devices, with assurance
> those addresses will not need to change the way public
> addresses might.

Ditto for PI.

S

Stephen Sprunk      "Those people who think they know everything
CCIE #3723         are a great annoyance to those of us who do."
K5SSS                                             --Isaac Asimov