RE: geo short vs long term? [Re: Geo pros and cons]

[Iljitsch van Beijnum <iljitsch@muada.com>]

On zondag, apr 6, 2003, at 20:27 Europe/Amsterdam, Christian Huitema 
wrote:

> The lessons are pretty clear. I remember a time, not so long ago, when 
> the shortest path from France to the UK actually was through 
> Princeton, and the shortest path from some parts of France to other 
> parts was through either Amsterdam or Geneva. The reality is that 
> network topology is primarily driven by network economics, which 
> themselves are only modestly driven by geography.
Yes, I remember those times too.  :-)  When I started out we were 
multihomed to a Dutch network and a British network and a lot of NL 
traffic over the British network went from London to Washington and 
back. Other European destinations were especially bad.

A 500 km detour (from Paris over Amsterdam or the other way around) is 
something we shouldn't consider out of the ordinary. In the US for 
example, only a tiny fraction of all traffic should be expected to stay 
within the same city or state. In Europe this is different for reasons 
such as language, and most countries have at least one interconnect 
location of national importance.

> In theory, it seems easier to lay fiber over short distances, but this 
> is one of those cases where the difference between theory and practice 
> is even greater in practice than in theory. In practice, you get into 
> things like right of ways that prevent crossing streets,
I'm not even close to worrying about streets...

> traffic pattern that are not local, satellites that don't particularly 
> care about imediate proximity,
Yes, satellites are a problem.

> or the simple fact that laying fiber accross a body of water is 
> simpler that doing the same accross a mountain ridge. Then, you get 
> competition between providers, and all sorts of non technical reasons 
> such as trust, business affinity, etc.
All of this can explain why European or Asian traffic flows through the 
US (to name an exampel), but that doesn't make connectivity completely 
arbitrary: even if European traffic flows through the US, it's always 
the east coast, and west coast for Asian traffic. (At least, as far as 
I have observed.)

It seems some of you guys are trying to convince me that geo 
aggregation can't work. I'm trying to convince you that it can and 
will. Rather than going over the same arguments again and again without 
convincing anyone, it might be a good idea to say what it takes to be 
convinced. For me, that would either be a fatal flaw in my draft that 
makes it impossible to get the necessary filtering off the ground in 
current BGP implementations, or something that shows that if geo 
aggregation is implemented in today's internet and new multihomers 
connect in a rational way, the savings in routing table size are 
negligible.

What would it take for you to agree that geographic aggregation can be 
a useful short to intermediate term tool?

> By the way, these non technical reasons apply also to various "virtual 
> addressing" schemes that propose to use a provider independent overlay 
> on top of a provider addressed network. Any kind of virtual 
> aggregation is much more likely to be centered on business relations 
> than on geography.
If you base aggregation on business stuff, you must renumber when 
business changes. I don't think that is going to work well. The sad 
truth is that despite the renumbering tools we have and are building in 
IPv6, renumbering is only getting harder, not easier. We know we can't 
trust the DNS and no global PKI yet, so most access restrictions work 
on IP addresses. That's bad when you have a few hundred globally unique 
addresses, but in IPv6 you can have many, many more hosts in a /48. The 
more hosts with globally visible addresses and the more subnets, the 
harder renumbering becomes.