[ppml] Revising Centrally Assigned ULA draft

[Paul_Vixie at isc.org]

this is a nice summary of an attractive viewpoint:

> 2) There is no guarantee of routability given.  If you want your
> /48 routed, you will have to find someone willing to route it.
> Presumably, they would charge you for it.  Presumably, they would
> charge you much less to route a /48 (or whatever) they assigned
> to you out of their PA space.

these costs would inevitably be transitive.  all router operators
would have to charge more for longer prefixes, whether as part of
settlement fees to their peers or as part of transit fees to their
customers.  the ratio of "how much this route helps me" to "how
much this route hurts me" is assumed to be worse for longer prefixes
(which take the same amount of router resources but offer a lower
number of possible destinations.)  the quality of the destinations
can't be considered in the global route-charge economy -- a /32 full
of spammers has to be presumed to be more valuable than a /48 with
a single root name server in it.

this is attractive in spite of its shortcomings simply because it
puts the cost of TE routes back on the businesses who use them, and
assures that if massive TE routing became the norm, there would be
enough capital in the economy to support the 10-megaroute equipment
that virtually all of us would need.  it allows someone who values
provider independence (either for high availability or anti-lockin)
to offer value in exchange, which is what a functioning economy
depends on.  this libertarian ideal of "use your dollar votes!" is
attractive almost everywhere it's first proposed.

but attractive or not, is it realistic?  can somebody show some
numbers whereby we can know what a small/medium business's "route
costs" would be for a /48, and how these costs would role up, and
how they'd affect typical peer-level settlement agreements at both
tier 1 and tier 2, and how they'd affect typical transit agreements
at tier 1 and tier 2?  i'm especially interested in how the economy
would support "cost aggregation" whereby it's not cheaper to take
your /48 to a tier 1 than to a tier 2, owing to quantity discounts
in those agreements which allow a tier 2 to "roll in" the costs.
(otherwise route charges just support more lockin than we have now.)

> Maybe the problem here is people are expecting ARIN policy to enforce
> their business policy so they don't have to give the customer the
> bad news that they are going to charge them an arm and a leg to
> route their small subnet.  Instead, they just try to keep ARIN's
> policy such that their customers can't get small subnets in the
> first place.

i think the problem is different.  here are four alternative problem
statements that could each yield the sentiment you're noting here:

	(1) ARIN's policies have historically attempted to balance cost vs
	    value of various prefix lengths, since although ARIN doesn't
	    operate routers for third parties, ARIN's members do.

	(2) a lot of us remember the 64MB RP limits of the early 90's and
	    aren't inclined to skate deliberately near/toward that edge.

	(3) a 10-megaroute router may become practical, but a 10-megaroute
	    DFZ doesn't feel like it would ever reach convergence, even if
	    we could afford the bandwidth to inflict that much BGP traffic.

	(4) nobody really knows how a route-charge economy would look in
	    practice, and so there's some fear that it could negatively
	    affect consumers and middlemen.  (sort of like electricity
	    deregulation in california during the Enron era.)

so: attractive yes, but practical maybe not, plz send out some numbers.