Invisible Hands, was Re: Multihoming sites and ARIN

[John W. Stewart III]

 > > you might be able to convince me of [market based distribution]
 > > for space that is already allocated.  but what do we do about 
 > > currently *un*allocated space? 
 > 
 > In a sence there is no unallocated space.  The IANA sits on all
 > of the space, they can delegate this space to their own close
 > personal friends as they see fit, or they can delegate to the 
 > registries, who then delegate according to whatever arbitrary
 > policies they see fit to employ.  :)
 > 
 > The question you ask seems to be "how much space should the IANA 
 > hold in reserve?" perhaps for meritorious use or perhaps for 
 > deployment to their own close personal friends. :)
 > 
 > Note that whatever amount the IANA picks you basically are attempting
 > to control the market supply in such a way that either 
 > artificially raises or lowers prices for IP space, this seems to me 
 > to be a bad thing, though I see that it might be a social good to 
 > reserve some amount of space, but I expect that amount should be 
 > "not much".
 > 
 > > do different registries on the supply-side of the market just 
 > > pick randomly and hope for uniqueness?
 > 
 > The role of registries is to record allocations, not make them.  
 > Lets have lots and lots and lots of competing registries, similar
 > to the IAHC proposal for domain names.  

the point about unallocated space is this.  if i have address
space that i was given 10 years ago, then it's pretty clear
that i can sell it as long as i find a buyer (independant of
all of these fun conversations).  however, if nobody that
already has address space wants to sell, and i want to buy,
then someone has to dip into the pool of unallocated space.
in what way do competing registries do this? first-come-first-
served?  that's fine if you divorce concern for the routing
from the address allocation, but i claim that that's irresponsible
unless you have a strong reason to believe that other forces
will ensure stability of the routing infrastructure.  and it
might even be irresponsible in toto .. whose to stop
allocator number one from claiming everything?

 > > if address assignment isn't [at least] loosely correlated with 
 > > topology, then a market-driven allocation system in a vacuum with
 > > respect to routing could end up destroying the routing
 > > system.  
 > 
 > Note that the current allocation schemes can destroy the routing
 > system and there are no methods to fix it.  For example, say
 > ARIN follows the InterNIC guidelines and dispense addresss space 
 > along topological bounds as it has, low and behold later this year 
 > we hit The Magic Limit for routing slots and we're full.  People 
 > can still go ARIN and ask for space but the space can't be
 > used on the global Net.  How do we go forward?  
 > 
 > Clearly, a market based scheme insists on making things work
 > (else why buy the slot or pay for the space)  While good intentioned,
 > it's not clear to me if procedural delegation can guarantee this
 > and it fact it is more likely that procedural delegation would
 > destroy the routing system than a market based scheme.

your final statement (starting with "it is more likely...") is
*very* strong.  and i think it's *very* debatable.  given the
current router technology and its configuration ability with
respect to inter-domain routing, i claim that the current "up"
status of the internet routing system is due to provider
aggregation .. an allocation policy which depends on the iana
and its three subordinate registries allocating space to
providers.  i don't think i could agree with your statement
unless the world were first turned upside down

 > > however, i claim that *if* i knew exactly what "full" meant for 
 > > a default-free routing table 
 > 
 > Then you would exactly know the supply curve in a market based
 > allocation :) BTW, how well do the current registries know this?
 > Don't they need to make the same sorts of assessments in making
 > policy on their allocation schemes?
 > 
 > I assert that a free market would determine the demand and supply
 > curves and an equilibrium point for a default free routing table.
 > Some Econ 101 attached at the end.  Note well, the example about 
 > creation of route slots by the introduction of new technology.
 > 
 > Further, I assert that any group, however well intentioned, and
 > regardless of the qualifications of it's Board of Directors, that 
 > attempts to manage a scarce resource will *not* find the 
 > equilibrium point, resulting in either scarcity or waste.

yes the registries may be creating an artificial scarcity.
numbers are numbers and policies can't change that.  but,
one could argue that the creation of an artificial scarcity
of ip addresses is better than the "true" scarcity in
routing table entries that would be created by trying to
inject "full + 1" routes into a default-free table.  in
other words, is artificial scarcity better than total
failure?  (note:  this is from the perspective of the world
as it is today and not a model where routing table entries
are sold and the implied configuration of inter-domain
routing is there)

 > > *if* i had the ability to control what routing announcements i 
 > > hear win (i.e., get a slot) and which ones loose (i.e., get dropped 
 > > on the floor), then it might be more possible to separate 
 > > allocation and routing 
 > 
 > I'm not sure of your point here, but I'll grant that this could
 > require some additional knobs in routers, though I kinda think
 > not.  Clarification?

right now access to a routing slot is first come first served.
fortunately, assuming you have a router with enough memory,
*all* get served.  however, when/if we start to reach a "full"
state (assuming that routers do something more gracefully than
roll over and die .. which actually requires some work given
the current longest-match routing), then first come first
served isn't good enough and, in the case of the selling-
routing-table-entries as your model proposes, you need to
ensure that the people who bought slots from you (and all of
your peers) don't get dropped on the floor

 > In the end, allocation and routing are already
 > seperate, i.e. ARIN allocates IP space, it does't route.

but they try to allocate in a framework which tries to
ensure routability .. they don't allocate in a vacuum

 > ARIN won't guarantee routability and won't guarantee that the routing 
 > table will not fall over.  Market mechanisms would.

i don't think that your last two sentences are true in
general .. i think that they assume other things (like the
selling-routing-table-entries).  they at least assume these
things in order for your approach to be practical

/jws

 > 
 > best regards,
 > 
 > -scott
 > 
 > Some casual background, this is Econ 101
 > 
 > The Demand Curve,
 > 	o	changes in price move you along the curve (at a high
 > 			price, low demand, at low price high demand)
 > 	o	changes in aggregate demand shift the curve
 > 
 > $ |
 >   |\     \
 >   | \  -> \
 >   |  \     \
 >   |   \     \
 >    -------------
 >         Q
 > 
 > The Supply Curve, 
 > 	o	changes in price move you along the curve (at a high
 > 			price, high supply, at low price low supply)
 > 	o	innovations in technology shift the curve right
 > 
 > $|     /     /
 >  |    /     /
 >  |   / ->  /
 >  |  /     /  
 >  | /     /
 >   -------------
 >         Q
 > 
 > The Great Truth
 > 	o	The "invisible hand" of a competitive market will find 
 > 			an equilibrium point, where supply *exactly* equals 
 > 			demand at some price, P
 > 			(at any other point, Evil Greedy Bastards arbitrage
 > 			 to exploit market inequalities and move to
 > 			 the equilibrium)
 > 
 > $ |     /  
 >   |\   /     
 >   | \ / 
 > P |  *       
 >   | / \    
 >    -------------
 >         Q
 > 
 > -------------
 >