[ppml] Free Market
Iljitsch van Beijnum
iljitsch at muada.com
Mon Aug 27 10:43:30 EDT 2007
On 26-aug-2007, at 19:59, William Herrin wrote:
> The topology of the Internet hasn't been hierarchical since the
> dissolution of the NSFnet and every passing day sees greater
> geographic dispersion of connections in the graph.
That doesn't matter. The point is that you get to express the set of
destinations that sit behind a particular connection with as few bits
as possible. So if you have a prefix that means "Europe" then you can
route that one prefix over your transatlantic link instead of 20000
prefixes for 20000 European destinations. In practice, it's not going
to be _that_ simple, but even if you get to route one aggregate and
you have to make 4000 exceptions, you've saved 80%.
> Its quickly getting
> to the point where even mere mortals can afford to lease transatlantic
But strangely, they can't afford to lease a fiber from their home or
business to the place where the transatlantic fiber sticks out of the
Still, it's incredibly inefficient to route all your traffic across
an ocean so few people do it if they have the choice.
> With today's traffic engineering requirements (and the economic
> pressure for better traffic engineering if only it was possible), the
> argument that aggregation can still work is just plain intransigent.
We can argue about that, but it's not relevant to the problem of
allowing millions of users to multihome. Those millions aren't going
to lease intercontinental fibers or even do complex traffic
engineering, they'll just get cable and DSL and type a router config
from a book if they have to.
> I read your geoaggregation paper. You don't give the multihomed end
> users or their ISPs any choice over which carrier moves their packets
> across the sea and you fail to address what economic structure allows
> the carrier who does move their packets across the ocean to get paid.
For outgoing traffic, nothing changes. For incoming traffic, you
don't get to influence much today anyway so not much changes there
either. The big difference (that I do talk about extensively) is that
with this aggregation, you get "cold potato" routing rather than "hot
potato" routing. Although this changes some things I don't think it's
a fundamentally different economic model. You still have peering and
transit the way we have today.
More information about the ARIN-PPML