[ppml] 2005-1:Business Need for PI Assignments

[Michael.Dillon at radianz.com]

> > In the end, I think that strict geographic allocations which
> > follow the topology of the network, are the only workable solution.
> 
> Make up your mind -- do you want strict geographic allocations or do you
> want addressing that follows topology?  Those are only the same thing 
for
> very simple end user scenarios.

I want strict geographic allocations that follow the geography of the
network. This is roughly similar to real-world geography except that
the distance between cities is not measured in straight lines because
the fiber routes are not straight lines. 

An analogy from the real-world is the underground metro system in
many countries. There are published maps which clearly show the
geography of the train lines connecting the various stations. 
However, one cannot determine the physical distance between stations,
even on the same line, because the map shows the network geography
and not the physical geography. Nevertheless, you can look at
a metro map for Moscow or London and determine whether a station
is in the east or the west of the city, whether it is near the center
or far from the center.

> Real world example: picture an enterprise with 40,000 sites in the US. 
> Should each of those sites be numbered from the local state/city pool? 

Should? I'm talking about "may" here. I want us to allocate geographic
addressing so that they MAY number each site from their local city pool
or they MAY number each site according to some other plan. I want to
create a situation in which the existing IPv6 address allocation plan
has to compete in the real world with a geographic addressing plan.

In the end, both systems may be successful because I believe that
smaller networks will prefer to use geographic addresses. This will
take enough pressure off the global routing table so that larger
enterprises can continue to use the current 2000::/3 addressing 
plan and announce their own global routes.

>  Please explain how _any_ addressing
> model other than PI makes sense.

Simple. The current PI addressing model requires an organization
to announce routes globally in order to obtain multihomed resilience.
The PI addressing model does not necessarily provide a way in 
which such route announcements can be aggregated, thus the current
PI addressing model cannot accomodate exponential growth in multihoming.

Since a geographic addressing model does provide for planned aggregation
at city and regional and continental levels, it can accomodate exponential
growth in multihoming when the multihomed organization buys services
from two or more ISPs in their city who support geographical addressing.

> The reason it's not being pursued is that (a) the topology does not 
follow
> geography today and (b) providers have motivation to make it so.

Topology does follow geography. So do a lot of other things. The network
exists in the real world. Obviously, one can pick nits with the fact that
all fiber lines do not exactly follow major highways, railways and sea 
lanes.
But that is irrelevant. The fact is that communications traffic broadly
follows the same pathways as real-world commercial traffic. New York
is well connected in both economic and network terms. 

> It'd make for a nice Ph.D. thesis though.

I'm not so sure there is a Ph.D. thesis in this but there is certainly
a Masters level project or two. It would be nice to see the results of
some serious studies on the issue. And it would be really nice if the
RIRs would, at least partially, fund some research in this area.

--Michael Dillon