Geographic v. topological address allocation
SEAN at SDG.DRA.COM
Tue Nov 11 05:25:09 UTC 1997
Either geographic or topological address assignment will work. What
works 'better' (I won't say 'best') depends on so-called externalities.
In today's Internet the sender's provider keeps all the money. The
sending provider wants the traffic off the sender's network as quickly
as possible and onto the destination's provider network. In this
case the goal is to minimize usage of the sending provider's network,
so you want address assignment based on the destination provider, so
you can find the nearest destination provider gateway from the source.
On the other hand, lets assume a different economic model. If instead
the sending provider had to pay the destination provider based on the
mileage (e.g. V&H coordinates) between where the sending provider gives
the traffic to the destination's provider and the final destination,
the sender would want to carry the traffic as close to the destination
as possible, assuming that its less expensive for the sender to use its
facilities than to pay to use someone else's facilities, and hand the
traffic to the destination's provider at the last possible gateway. In
this case, the goal is to minimize mileage charges on the other provider's
network, so you want the address assignment based on geography, so you
can find the closest sending provider gateway to the final destination.
The clunker in most models is traffic flows in both directions, and
money also exists at both ends. Neither model is very good at global
optimization of traffic or the money. Even having a single payer and
single provider, such as was used in the NSFNET, resulted in sub-optimal
use in some cases due to yet other externalities.
What's interesting is as a provider's network reaches different sizes,
different types of optimization become more desirable. And, yes, both
styles of addressing can be used to make it difficult for competitors
to enter the market.
More information about the NANOG