The Year to Peer Is Clearly Near


Six months have passed since Excite@Home Corp. service finally flamed out, and — like saplings nudging through blackened earth in wildfire country — new signs of healthy life are already emerging.

One such sign is the steady advance of "peering" agreements between cable providers and other, nonaffiliated Internet-service providers (ISPs).

Born in the Internet's go-go days, peering is one of the few ideas of that era that didn't wind up in the dot-com junkyard. To peer, in this sense, is to agree to link up, not to stare. You peer with
someone, not at them.

Specifically, peering is what occurs after a cable operator that offers broadband Internet service realizes that it is exchanging so much Internet traffic with someone, on such a regular basis, that it's probably time to link to them directly.

The moment it becomes more expensive to pay someone to haul traffic to an entity than it is to link to that entity directly is the moment when peering discussions begin.

ISPs, regardless of breed (dial-up, cable modem, or DSL) pay in the range of $150 per megabit to move their customers' traffic to its destination. (People with less traffic pay more; people with more traffic pay less.) The megabit calculation is based on peak traffic times.

Example: Say you're an ISP, and you've put the mechanisms in place to examine what tech people call "network flows." That means you've purchased software that automatically monitors the traffic through your core routers. With it, you see how and where the data to and from your network is moving.

Over a period of time, you realize that you're consistently sending lots of traffic to another ISP, and receiving similar amounts of traffic from that source. Maybe you're sending 50 megabits per second, at peak times, to them, and they're sending 50 mbps, at peak times, to you.

In this very simplified example, you're both paying different transit providers — these are generally the long-haul companies, like AT&T Corp., WorldCom Inc., or Sprint Corp. — to move those packets to each other. You're paying a lot, actually: At $150 per megabit, each of you is shelling out $7,500 per month to get your traffic to and from each other.

Wouldn't it be cheaper, you wonder, to make a direct link to each other?

Doing the math involves three main cost points. First is the price of getting your traffic to a mutually agreeable exchange point, called a "NAP," for "Network Access Point." A NAP is a place — usually a nondescript room in a nondescript building — populated with racks of routers. Once you're there, you'll need to rent some rack space and buy a port into the matrix of gear there that moves your customers' data to and from the wider Internet.

If the math works — and if it looks like a peering breakeven point is within reach — the next step is to figure out where to do it. In cable terms, this is the tricky part right now. The demise of Excite@Home caused all of its former constituents to build their own regional and national backbones, which don't necessarily intersect.

Most MSOs are glad to have control over their networks. (Mostly, they're glad to have the whole @Home inferno behind them.) However, in @Home's wake, all previously affiliated MSOs are lacking the backbone that at one time was
their peering agreement.

This mutual access among MSOs doesn't matter so much right now. There just isn't a lot of activity in advanced IP applications that could dramatically benefit from a mutual, nationwide backbone, like the former @Home network.

As services like voice-over-IP telephony develop, though, peering agreements among MSOs will make more and more sense. Recall that one of the benefits of cable-delivered VoIP is the ability to handle a telephone call without ever touching — which is to say paying — to hop on and off of the public switched-telephone network.

At the moment, Cox appears to be the most assertive in MSO-to-MSO peering. So far, it maintains or plans to maintain a presence in four NAPs, which are located in cities that map with its seven regional clusters. Other MSOs say that peering is on their radar.

As the discussions advance, haggling points will probably revolve around traffic sizing: the "but I'm sending more to you than you're sending to me" stuff that can temporarily stymie progress.

In the end, though, the benefits of peering will far outweigh the economics of maintaining networks that don't touch each other. Peering saves money — for Cox, which averages about 6 gigabytes of monthly traffic, that's between $100 and $300 per megabit, per month.

These days, anything that saves money is good. That's why peering is poised to become a big part of broadband Internet discussions over the next few years.

Questions? Suggestions? Write to Leslie Ellis at