In the past, some providers have specified peering requirements such as "must use a Cisco 7000 with SSP". If you are concerned about congestion, you should specify a performance requirement covering congestion, not some internal architecture design of the other network. How the other network chooses to solve the congestion problem should be up to them. By saying you should treat the other provider as a black box I mean you should be able to tell if the other provider is meeting or not meeting their performance requirements with examination of how much they are paying for their circuits, what type of routers they are using, or how they design their internal network. On Sun, 14 January 2001, Paul Vixie wrote:

...

... I think interconnection agreements should be based on the point of interconnection. When you delve too much in how the internals of other providers' networks work, I think you are always going to run into problems. I think it is best to view other provider's networks as a black box.

I don't agree. Peering is a business relationship transcending locations. The best peering agreements I know of (e.g., mine) specify growth terms so that traffic won't eventually encounter congestion while trying to get from one to the other or back again. Some forms of growth involve additional peering locations, and some do not. A peering agreement which was based on the point of interconnection would be far less useful in avoiding congestion.