[arin-ppml] Abandonment of 103/104

michael.dillon at bt.com michael.dillon at bt.com
Wed Dec 23 16:46:36 EST 2009


> Is there a good reason that ARIN doesn't pick 1 to 5 hubs in 
> a province/state and assign a /32 or larger to them depending 
> on their Internet population and then make sub-allocations 
> from these region prefixes to smaller entities in that region?  

When I suggested something like this, the feeling was that
the IETF would have to bless it first. My idea was for the
IETF to give each RIR a block that was big enough to carve
out one suballocation region for each of the world's major
cities that are within the RIR's region. If possible, those
city allocations would be aggregated into clusters of cities
that were close to each other, for instance New York and
Montreal and Boston would be in the same cluster, San Francisco,
Las Vegas, LA, and Tijuana in another. That sort of thing.

The basic rationale was that the world's physical network
links tend to converge on a small number of major cities,
somewhere between 100 to 1000 cities depending on how fine
you want to slice it. And the small organizations that want
to multihome typically either want to stay in the same
building and change local providers, or move to another
site in the same metro area, possibly with a new provider.

There's a lot more to it than that, but if the RIRs would
implement such a scenario, it would add 1000 new city-aggregate
routes to the DFZ, and would give end users the option of
PA, PI or CA (City Aggregate) addressing. The existing PA
and PI allocations would be unchanged, and no doubt, some
providers would ignore CA altogether and it would be more
popular in some cities than in others.

If we did implement this, then there is no longer any
good reason for the ITU to call for country allocations
because, for instance, half the CA address users in
the Strasbourg, France aggregate would be located
across the river in Germany. Maybe Tijuana would be
part of a San Diego aggregate.

> Huge amounts of the potential global routing tables could 
> then be aggregated regionally with still gaining the ability 
> allow small local sub-allocations for small entities without 
> much impact to the global routing.  Then from the regional 
> aggregations could be broken up locally by ISP.

Exactly my point. Perhaps the time has come to take another
serious look at it. As I recall last time, there were no
substantive technical issues with doing this. The biggest
problem people had was that it was not guaranteed to work
unless imposed from above, and then people attacked that
strawman of "imposed from above". I always viewed it as
an experiment of sorts where we make the address space 
available to those organizations who want to try and
make City Aggregate work. Even if it only works in half
the major cities of the planet, that is 500 cities that
are not spewing new entries into the DFZ.

In my opinion, there is no grand solution to DFZ growth,
just a bunch of small improvements that slow DFZ growth
and when all of these small improvements are combined,
they keep DFZ growth manageable. Maybe in 10 years from
now the RRG and IETF work will come to fruition and we
will have a grand solution, maybe not.

> And I'm fine with the idea that if the entity moves out of 
> the regional then they do have to re-address or if they open 
> branches in other regions, those branches get regional 
> addressing based on their location.

Yes. There is not point in asking an improvement to be
perfect and to deliver 100% of the desired benefits.
If it works for over 80% of the cases, then it is worth
doing.

--Michael Dillon

P.S. I will work with anyone who wants to hone this idea
into an Internet draft.




More information about the ARIN-PPML mailing list