Strikes me that the terms "diversely routed" and "geographically diverse" can be interpretted wildly.

You catch on quick. Who gets to do the audit? Other than the 'public' interface between the providers, I don't think it is wise, or appropriate to get into the internal design or business plans of competitors. You really have to treat the rest of it as a black box. It hard enough to predict what technology I'll need to deploy in my network without having by hands tied by an agreement with a competitor. Thou shall use only dedicated, fully-meshed, terrasterial clear-channel lines sounds great until I discover the best way I can reach timbuktoo is over a ATM satellite link. DRA doesn't have large Web farms, but people build public libaries in the darndest places. I'm always amused by providers that have peering requirements their own networks, or existing 'peers' don't meet.

For most networks, I don't see a real problem dropping a couple extra pairs of DS3s (assuming they were only at two NAPs to begin with).

There is a partition problem. Assume there are six points, and each provider chooses four. Provider 1 could choose 1, 2, 3, 4. Provider 2 could choose 3, 4, 5, 6. Although each provider is at four points, they only share two in common (3, 4). Which provider alters their business plans to connect to eight places just so they have four common exchange points?

If UUNet requires a DS3 interconnect at LINX, MaeWest, MaeEast and say (for giggles) Australia -- AND a diversely routed DS3 backbone in the U.S. that takes on an entirely different meaning than the above.

Don't giggle, DRA has an office in Melbourne. We don't have enough traffic to justify multiple DS3/E3's to Melbourne, but looking at UUNET's international map, neither do they. Every provider has their own strengths and weaknesses, they are never exactly equal. Getting into the internal network design of your competitor is just asking for trouble. PSI and Sprint engineers used to battle over who had the better backbone design. Neither ever convinced the other they were right. I don't think its a good idea to require every backbone follow the exact same design. We may have different 'major' routes than other providers, and choose to deploy our 'big' links differently. That's one of the strengths of the Internet, not a weakness. My design methodolgy may be good, or it may be bad; but the design diversity between providers hopefully means a "group think" mistake won't kill everyone. -- Sean Donelan, Data Research Associates, Inc, St. Louis, MO Affiliation given for identification not representation