Commentary: ISPs are class-conscious

By Rob Batchelder, Gartner Analyst

Peering is a type of gentleman's agreement for the Internet whereby service providers of the same class enter into handshake agreements. Unfortunately, peering also has the same vulnerabilities.

A large web of peering agreements holds the Internet together. Beginning with the largest backbone carriers, networks of comparable size agree to carry each other's traffic in a reciprocal manner.

We can't overstate how valuable peering relationships are at the top of the food chain. Peering at that level gives networks access to very high bandwidth connections and very high throughput. Not surprisingly, every company tries to climb the social ladder in their peering relationships, but the only way to do so is to prove that they consistently--and the emphasis is on "consistently"--have the volume of traffic to make them equal to the companies they aspire to peer with.

PSINet's size gave it a privileged status and allowed it to peer with the largest of backbone providers. However, as with gentlemen's agreements, if one of the parties falls on hard times and moves to a lower status, those agreements come unraveled. That's exactly what has happened to PSINet. Its traffic has fallen off, and it no longer "qualifies" to be in the uppermost tier of peering arrangements. Cable & Wireless (C&W), it seems, tried to work this out with PSINet but ultimately decided that it had had enough.

Unlike the regulated telephone companies, Internet backbone providers are not obligated--other than contractually--to peer with others. As in the case of C&W, they can end those relationships when one party fails to live up to its commitments.

That leads to a more interesting topic. With the breakup of the Bell system, the public telephone carriers implemented ways to cross-charge each other for traffic that transits their networks--an issue generically called "settlement." To make settlement work, all Bell companies had to implement an extremely elaborate traffic monitoring and financial accounting system--a monumentally complex, Byzantine system. Imagine cross-charging for billions, if not trillions, of events and relying on automated methods for reconciling irregularities that can occur in such a network.

When the Internet was built, its creators knew the complexities that settlement would introduce to that type of network. As the Internet transformed from a government and academic network into a public resource, its overseers made a conscious decision not to attempt a settlement system for the Internet. Doing so would have placed a huge burden on those who wished to connect to it and would have greatly slowed its growth.

	See news story: Net blackout marks Web's Achilles heel

Nevertheless, it happened that very early in the game the large carriers had to arrive at some understanding about carrying each other's traffic. Since neither party would be directly compensated for carrying that traffic, the convention became that networks of equal size would act reciprocally. Hence peering was born.

If a large peer drops from that category, the other peers have to decide individually or among themselves whether to keep their peering relationships in place. In that regard, the Internet is vulnerable. It does not follow, though, that events such as the demise of PSINet place the Internet on the verge of cascading failures. The situation is so rare--that is, what happens when a large network goes away?--that few people have thought about it.

More should, however. In the United States, state and federal regulations provide a safety net under the telephone companies, but the Internet has no such safety net. Therefore, enterprises should be highly concerned about the financial viability of their Internet service providers--and they have every right to interrogate a provider about its financial situation and peering agreements.

(For related commentary on how to select an Internet service provider, see TechRepublic.com--free registration required.)

Entire contents, Copyright ? 2001 Gartner, Inc. All rights reserved. The information contained herein represents Gartner's initial commentary and analysis and has been obtained from sources believed to be reliable. Positions taken are subject to change as more information becomes available and further analysis is undertaken. Gartner disclaims all warranties as to the accuracy, completeness or adequacy of the information. Gartner shall have no liability for errors, omissions or inadequacies in the information contained herein or for interpretations thereof.