Peering Agreement Dispute Between Level 3 and Cogent

By Eric Goldman

Peering agreements rarely get much attention, even though they are the Internet’s infrastructure. Through peering agreements, Internet access providers (IAPs) agree to exchange packets directly with another IAP. These exchanges are usually for no money with the assumption that the data flowing between the two IAPs will be roughly equivalent. If, in fact, traffic is roughly equivalent, it’s cheaper to barter than to charge for packets.

This assumption animates a bedrock and venerable principle of Internet data flows–that there should not be marginal costs to trading packets. This principle is false on its face, as in fact each party to a peering agreement incurs both non-trivial marginal and fixed costs to maintain the connectivity. It’s not that hard to imagine a different Internet evolutionary path where each IAP charged those marginal costs to other IAPs, in which case undoubtedly this variable charge would be passed through to consumers.

It’s relatively rare for a peering arrangement to blow up. The last well-publicized event was a dispute between Cable & Wireless and PSINet in 2001.

Then, last week, Level 3 and Cogent mixed it up. Level 3 complained that it was larger than Cogent, so Cogent should have to pay Level 3. Level 3 stopped accepting Cogent’s packets, so some Level 3 customers could not reach some Cogent customers. Not everyone was affected because many major IAP customers maintain multiple/redundant IAP relationships so that, even if one goes away, often the packets can route around the outage. Nevertheless, Level 3’s actions led to a fairly significant outage across the Internet.

The companies have kissed and made up for one month. However, Level 3 continues to insist that Cogent will ultimately need to pay up. This dispute may not be over.

This dispute showed the relative fragility of the current Internet infrastructure. If one major player decides to change the economic rules, (a) some customers will be cut off, and (b) there could be ripple waves that could easily change the default packet processing business model from barter to for-pay. This has created some fertile ground for Congressional grandstanding, and Rep. Boucher is hinting/threatening to use his Congressional powers to unilaterally fix the Internet.

I’m not inherently opposed to a model where we pay by packet. Economists would generally favor this because per-packet pricing would more accurately signal the true marginal costs of communication, encouraging communication at an economically efficient level. For example, if spammers had to pay per packet, the working theory is that spam would radically decrease.

However, the current infrastructure–and resulting lack of per-packet charges–has been IMO a material/essential factor contributing to the proliferation of Internet communications. We take it for granted that we can communicate on the Internet without incurring some annoying and minor marginal cost, and this psychology has enabled the Internet to weave itself deeply into our society. Perhaps we’ve had it good for a while, and all good things must come to an end. On the other hand, it would be a shame if we lost those benefits simply because one IAP freaked out.

UPDATE: The Washington Post quotes a bunch of putatively anti-regulation people thinking this is a situation that warrants regulatory intervention.

UPDATE 2: Notice the ads on the left. Because of this post, I’ve been getting ads from Cogent promoting its offer to Level 3 customers giving them 1 year free if they switch to Cogent.

UPDATE 3: Cooler heads have prevailed, the parties have settled their dispute, and it appears that the parties are generally exchanging packets without payment.