[ppml] Address Space versus Routing Slots

Michael.Dillon at btradianz.com Michael.Dillon at btradianz.com
Tue May 9 05:00:14 EDT 2006

>   Aggregation is NOT the solution.  The need for aggregation is proof
> that we have failed to develop a scalable routing architecture.

On the contrary. The need for aggregation is proof that
we have designed a hierarchical routing architecture
that is based on aggregation. This started when CIDR
was introduced and has been honed a bit since that time.

Hierarchy is known to be the KEY STRUCTURE in creating
scalability in networks. You may feel that you can create
a more scalable network by using a different technique
for creating hierarchy, i.e. hierarchical locators with
a flat identifier space, however it is overstatement to
say that aggregation is a sign of failure.

>  Why
> are so many people determined to preserve this failure? It makes no
> sense to me.  Why not look for a scalable routing solution that can
> be depolyed without requiring aggregation?

I support your effort to find a scalable way in which
to split locators and identifiers in IPv6. However, at
the same time I believe in free competitive markets. That
is why I also support the implementation of geo-topological
addressing in IPv6. The two solutions are different in many
ways. Geotopo addressing can be implemented sooner with no
new protocols or router code changes. But locator/identifier
split holds the promise of greater scalability (bigger sized
network). Geotopo is closer to the current way of doing
things which means that it is more likely to gain traction
sooner and therefore provide benefits of reducing the
global routing table size. But, if the locator/identifier
split can be implemented and eventually supercede geotopo
then that is not necessarily bad.

It is called evolution. Along the way to the end game
there are many solutions that are right for their time
but which are ultimately discarded. They are just as
essential as the end game because they help us attain
that goal.

Quite frankly, given the amount of work needed to design,
implement and deploy a locator/identifier split, I would
have thought you would welcome the breathing time given
by a shorter term solution. One thing that would really
kill locator/identifier split is to release it into the
wild before it is completely figured out.

And in the end, ARIN is a place for ADDRESS ALLOCATION
policies which includes some form of geotopo algorithm.
IETF is the place for new protocol design.

Going back to the subject line...
How can 2005-1 be implemented in a way that conserves
future routing slots? Allow for future aggregation.
The easy algorithm is to take a large reserve of IPv6
addresses and allocate west of the Mississippi from the
bottom-up and east of the Mississippi from the bottom down.
For the purposes of this algorithm, we should consider the
Manitoba/Ontario border to be the northern continuation
of the Mississippi and Nunavut should be in the west.

The tougher algorithm, but still within ARIN's abilities 
to implement, would divide the region into 8 to 10
sub regions trying to focus on the centers of city
clusters which have richer connectivity within the cluster
than without. CAIDA could help with this determination
although I suspect that a brute force technique of simply
aggregating neighboring LATAs would work well enough.
Unfortunately, a collection of LATAs has a hard boundary
which really is against the geotopo principle of using
center points of node clusters. The city is the archetypical
centerpoint of a node cluster because it has very rich internal
connectivity compared to external connectivity and the 
external connectivity tends to be directed to other
city clusters.

In order to do a full implementation of geotopo addressing
using the 5,000 or so cities with population over 100,000
we would need the IETF to agree to set aside 1/8 of the
IPv6 address space for geo-topological addressing and
an allocation plan acceptable to geographers and economists.

--Michael Dillon

More information about the ARIN-PPML mailing list