Toll Roads May Be the Future Route of the Internet
I thought it would be a simple matter when, two weeks ago, I started looking for a map of the Internet. I just wanted a diagram that showed the main pathways of the Net, the fiber-optic cables that crisscross the country carrying World Wide Web pages from San Francisco to Dallas and electronic mail from Los Angeles to New York. I thought that if I knew what the Net looked like, I might be able to determine if we are about to run out of network capacity, as some pundits have recently predicted.
I still haven’t found a map. In fact, I no longer think one exists. It turns out that the companies that operate the main arteries of the Internet consider detailed information about their piece of the network to be proprietary. It’s a strange situation, almost as if states refused to disclose details about how many highways they have and where they are located because it was considered competitive information.
Nonetheless, from talking to the engineers who are building and maintaining the Internet, I’ve been able to piece together, if not a map of the Net, at least an outline of its shape. Enough to determine that the roads that make up the Internet are almost running at full capacity and traffic jams are becoming increasingly frequent. And that, although this condition will exist only until the infrastructure is improved, the tremendous growth in Internet traffic is prompting fundamental changes in the underlying economics of the network.
Increasingly, we’re going to hear about “settlements,” whereby small Internet providers pay larger ones in order to traverse their networks. This may be the death knell for many of them. To understand why, you need to understand the Internet’s structure. Which brings us back to the missing map.
A few years ago, a map of the Net would have been easy to draw. The Internet’s backbone--the main pipes that carry data across the country--was owned and operated by the National Science Foundation, a federal government agency. This backbone, capable of carrying 45 megabits per second, connected a few dozen important scientific sites like the San Diego Supercomputing Center, UC Berkeley and the Massachusetts Institute of Technology.
Also connected to the backbone were regional access providers like CERFNet in Southern California and NEARNet on the East Coast. These regional networks provided Internet service to thousands of smaller sites and simply fed data that needed to go across the country to the NSF backbone.
But in April 1995, the NSF, having decided to turn the Internet business over to the private sector, turned off its backbone, and suddenly mapping the network became a much murkier business. Instead of one backbone, there are now almost a dozen, and instead of a few regional providers, there are now a few thousand small-time Internet providers.
The two largest backbones are operated by SprintLink and InternetMCI, but backbones run by companies such as AlterNet (which is partnered with Microsoft) and Advanced Network & Services (which was bought by America Online in 1994) are becoming increasingly important. All of these backbones are made up of what are known as DS-3 lines, capable of handling data at a speed of 45 megabits per second.
But nobody knows the total number of DS-3 lines being used to carry Internet traffic, or exactly what cities they run between, which makes it hard to say how much traffic the Internet can really support. And an even more important implication of having multiple backbones comes from the necessary interconnections. After all, if my computer is connected to SprintLink, and Microsoft is connected to AlterNet, how can we send messages back and forth?
The answer lies in NAPs, or Network Access Points. These are locations where the big Internet providers get together and exchange data. For example, Pacific Bell operates an NAP in Oakland that some 20 Internet backbones and major service providers connect to. In the scenario above, my message to Microsoft would be sent via our connection to SprintLink from San Francisco to Oakland. Then, at the PacBell Network Access Point, it would be switched onto AlterNet’s network and carried up to Seattle.
This works fine if everyone is exchanging equal amounts of data. But in reality, the smaller players end up dumping most of their traffic onto the big networks, while the big players end up handling most of their own traffic. Not surprisingly, the big guys see this as unfair.
They have ignored the matter up to now because, as one Sprint engineer put it, “the traffic volumes were so small it just looked like a rounding error.” But as volumes swell, and as network capacity grows tighter, it seems very likely that a system of payments will be instituted, whereby small providers have to pay for the privilege of exchanging data. That may sound reasonable, but expect to hear screams of outrage. The Internet provider business carries very low margins, and the introduction of settlements is likely to drive some of the smaller fish right out of the pond.
Nonetheless, as Bob Collet, president of the Commercial Internet Exchange (CIX), points out, “the telephone system has relied on a system of settlements for years. Sure, Internet folks are used to getting things for free, but that’s just not going to last.”
Pushpendra Mohta, executive director of CERFNet, agrees. “We’re large enough that we don’t need to pay settlements, but you will probably start seeing smaller regional providers being asked to pay settlements this year.”
This may very well be the most important outcome of the approaching Internet capacity crunch. Contrary to what some analysts have proclaimed, I don’t believe that the Web will collapse from lack of bandwidth. By next year, companies like Cisco Systems and Ipsilon Networks will have developed the equipment necessary to run Internet backbones at speeds as high as 622 megabits per second. But until then, watch out for an increasing number of toll roads.
