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

[Paul Vixie]

> > in other words, the definition of "routable" depends on who
> > you want to be able to exchange packets with.
> 
> ... The question of what "routability" is is not one that I'm interested  
> in.  We know what this means today, massaging the definition to fit a  
> particular purpose can only lead to suboptimal results.

we do not know, for the purpose of understanding and implementing a ULA
policy, what "routability" means today.  your lack of interest in it is
not going to advance the debate as to whether ULA is a good or bad policy.

> > "local" and "routable", not so much so.
> 
> Ok, if that makes you happy:
> 
> Routable address space: any block of global unicast address space that when
> announced through or by an internet service provider, allows the holder to
> receive packets addressed to the addresses in question from all possible
> sources connected to the internet without additional effort.

"the internet"?  let me quote this again in case you missed it the last
two times:

	>> But what *IS* the internet?
	> It's the largest equivalence class in the reflexive transitive
	> symmetric closure of the relationship "can be reached by an IP
	> packet from".		--Seth Breidbart

smaller private connectivity domains would only depend on uniqueness among
their participants.  are you trying to make a definition of "routable" that
depends on which connectivity domain the observer is actually in?  or would
it be enough to say "the connectivity domain that ARIN's members all share"?

> ULA fails this test because it falls outside the global unicast block  
> and because announcing it to one ISP isn't enough to receive packets  
> from all over the world because people will filter.

so the routability of an address block can also depend on filters?  (and
perhaps on firewalls?)  and is an address block that's "routable" today
capable of being "unroutable" tomorrow?  and if it's "routable" according
to network X, can it be "unroutable" according to network Y?

i can't imagine how a policy with these dependencies would be implemented.

which is precisely the point i'm trying to make.  the RIR system can
guaranty uniqueness among RIR allocations, but can make no assertions
either way as to "routability".  since any definition of ULA depends on
a definition of "routable", i think this slope is too slippery for us.