[arin-ppml] Creating a transfer market for IPv4 addresses

Milton L Mueller mueller at syr.edu
Tue Jun 17 18:14:27 EDT 2008


> 
> I think it is rather difficult to internalize the routing externality.
As
> Geoff Huston says there is no routing economy.  The problem is that
there
> is a real cost of consuming a routing table slot (RIB), a forwarding
table
> slot (FIB), and CPU to deal with routing instability with a reasonable
> amount of convergence time.

Jason, 
I think the issue of a routing slot fee or market is a very interesting
and complex one. I've had some very preliminary discussions of it with
Geoff Huston. At the time he seemed to believe that because someone in
1996 (Yakov Richter, I think) tried to write one paper about it and
didn't get very far that it can't be done. But then, that was three
years ago and in 2007 Huston became the author of APNIC's proposed
address transfer policy, didn't he? 

Right now we are talking about address transfers, not a routing market.
I don't see the two as linked. 

I think Jon Radel hit the nail on the head. "I would expect the price of
addresses to reflect their quality." If the routability of a tradeable
address block is in question, the willingness of people to buy them will
decrease proportionately. I don't think anything else needs to be said
about that. If no market develops because of those uncertainties then no
harm done.

Lots of expensive assets have lots of uncertainties associated with
their value and functionality. Think of buying a home, or even renting a
apartment. Who knows what unpleasant little surprises await you. And
yet, people buy them. And we're all better off because they can be
traded. 

Getting back to Bill Manning's question whether anyone out there really
wants to buy address blocks. That discussion is getting a bit silly. We
need to explore and discuss the merits of authorizing private address
transfers; I am not going to spend another minute humoring the pretense
that no one is interested in transferring addresses. 

I am justified in doing that because, if anyone really believes that
there are no substantial players out there willing to trade address
space for money, then that person shouldn't worry about instituting the
transfer policies. If no one trades, the transfers can't possibly do any
harm, can they? 

You can't have it both ways. EITHER address transfer policy proposals
are "shock therapy," the end of the RIR world as we know it, a form of
stealth privatization that will lead to rampant speculation, the
collapse of routing tables, the fragmentation of the Internet and the
heartbreak of psoriasis -- OR they are pointless exercises that no one
is interested in. Take one, or the other. Then we can have a discussion.

> -----Original Message-----
> 
> Scaling the RIB memory, the FIB memory, and the routing processor CPU
have
> costs, not just in dollars, but also in size, weight, power, cooling
> requirements, etc.  Assuming we can reasonably predict the growth of
the
> routing table, assuming the growth rate doesn't change too suddenly,
and
> assuming we can reasonably estimate the costs of scaling a router, we
will
> still have a problem of distributing the cost.
> 
> The distribution problem is as follows:
> 
> 1. Identify all of the ASes that receive full routes
>    1A. Alternatively, identify all of the ASes that cannot function
>        without receiving full routes
> 
> 2. Identify the number of routers in each of these networks that
receive
>    full routes
> 
> 3. Identify the life span of these routers as a function of the number
of
>    Internet routes, the number of internal routes, the number of
places
>    these routes are learned, the BGP architecture (flat iBGP vs
hierarchy
>    with some number us clusters / sub-ASes being aggregated at a given
>    level of the hierarchy), and the amount of routing churn
> 
> 4. Recover enough cost by charging per prefix in the projected life
span
>    for all networks that carry full routes to upgrade all of their
routers
>    that carry full routes
> 
> 5. Distribute this money equitably to all the networks that carry full
>    routes
> 
> To say it another way, should an ISP charge an end user to place a
route
> in the routing table, and then share a portion of those proceeds to
every
> other network that carries that route?  Should that portion be
determined
> by the cost of what upgrades cost?  Should ISPs with more routers (and
> thus a higher upgrade cost) receive a greater share of the proceeds?
> 
> Also it seems there is not a strong correlation between routing churn
and
> Internet table size, but rather only a few networks are responsible
for a
> majority of the churn, so this seems less predictable.  Should ISPs
also
> charge for routing churn and share a portion of those proceeds to
every
> other network that see the churn?
> 
> What about multi-homed end-sites that receive a full routing table?
Is
> this only a benefit to them to use shorter paths, and therefore they
> should carry the burden, or should they receive a share of the portion
of
> the proceeds from charging for routing slots?
> 
> __Jason
> 
> >
> > > -----Original Message-----
> > >
> > > The arrangement of the address space (in terms of numbers of
> > > blocks, shape of said blocks, and fit to existing holders blocks)
are
> > > all factors which affect the value of some space to a given
holder.
> > > i.e. 16 random /28's may not have the same usefulness as a /24)
> > > Most of this consequential to the externality of having to route
> > > the address space before being able to use it for most
applications.
> > > As there is no market for global table routing slots, and yet a
very
> > > real limit on their availability, this externality is extremely
> > difficulty
> > > to assess and creates a continuously iterated prisoner's dilemma
> > > situation with unknown equilibrium points.   Market failure in
> > > this case can result in a truly dysfunctional Internet.




More information about the ARIN-PPML mailing list