B O T T O M L I N E

TRAFFIC SENSE

Local Internet routing could be a way to ease congestion and lower costs, but consumers need to lead the charge.

by Joel Snyder

"What if we built a back door?" I was having lunch with a friend who owns the biggest Internet service provider (ISP) in the Tucson, Ariz., area. We were talking about how congested the main network access points (NAPs) are. The NAPs are the points in the Internet where the big boys link up. In effect, they have become the backbones of the Internet--the place where your packets go to jump from one major ISP to another. Think of the NAPs as big meeting places, where everyone can sit around the same table and share data. Congestion in the NAPs affects almost every Internet user, and is a big part of what makes the system so sluggish at times.

Only a handful of Internet service providers are directly connected to one or more NAPs. These are called "first-tier" providers, and they include the largest of the large: Alternet, BBN, MCI, Sprint, and a few others. All the other ISPs must buy service--directly or indirectly--from at least one of these first-tier providers. That ensures them a path across the backbone--one of the NAPs--to all other ISPs. So if you're using a local ISP, it's a good bet that it has a link to one of the first-tier providers. There are even third-tier providers who buy connectivity from the second tier and resell it.

The problem is that the NAPs have become horribly congested. The Internet market is massive, and no one player is dominant. That means the odds are pretty good that a large percentage of your electronic packets of information (your e-mail, for example) will have to pass through one or more NAPs as it travels to its final destination. Whenever packets go through a congested NAP, there's a chance of losing data, which translates into poor response time or a plain inability to connect. The congestion and data loss have become so bad that some first-tier ISPs have built private links so that packets from their customers don't have to go through the NAPs, but can travel swiftly over dedicated lines.

Shortening the electronic journey
Look at a city like Tucson, which has at least a dozen T1 links coming in from first-tier ISPs. Most of those links go to second-tier ISPs, who then resell access to the Tucson community as dial-up accounts, dedicated lines, Web hosting, and so on. The rest go to organizations large enough to need their own first-tier link, like the University of Arizona.

This creates a peculiar routing problem, with two Tucson ISPs shipping their packets via Los Angeles, San Francisco, Denver, and Dallas in order to end up across the street or down the block. It's an expensive way to make a lunch date. And that brings us to my question: Why not build a back door?

The idea is simple: Packets going from one side of Tucson to the other should stay within the city. Build a shortcut route among local companies so that the expensive and congested links to NAPs are bypassed.

This has three benefits. First, local traffic doesn't contribute to existing NAP congestion. That makes us good citizens on the Net. Second, local traffic doesn't go over expensive lines to the first-tier providers. That saves money because we can either have slower lines or put off buying faster ones. Third, performance is better. Electrons and routers move quickly, but there's a big difference in the delay (called latency) you experience going across town as opposed to traveling hundreds of miles.

There's even a fourth benefit: better communication and cooperation within a city. Even though many of the companies that might participate in such a scheme are ISPs who compete with one another for customers, there are benefits to the community in having good personal and professional relationships among ISPs. It's always a good idea to be on good terms with your neighbors.

Nothing to lose
So it's a win-win situation. What are the drawbacks? Well, if it's done right, the answer is none. The biggest problems are technical: These back doors have to be engineered properly--something that may take considerable time and thought.

The architecture is simple: Imagine a central point, with lines radiating from the center to each of the participants. Everyone who wants to participate has to get a connection to the central point, and that's where the packets will be switched from one local Internet user to another.

These back doors are inexpensive to put together within a city because the local exchange carriers (LECs), the companies that own the telephone system in each city (such as US West), offer favorable rates for this kind of network. Most companies who would build these back-door routes already have a connection to the LEC frame-relay network. That means adding a connection to a central hub is dirt cheap. For example, in Tucson, it costs less than $100 per month to add a T1 circuit to an existing frame-relay connection.

All it takes to set up a mini-NAP is for one of the participants to act as the hub site and everyone else to order a circuit to the hub. There's a slight inequity in that the hub's resources are used more heavily than those of the spokes. On the other hand, the hub has better access (one hop instead of two) to all other participants, so there is a certain technical advantage to volunteering.

There would also be a reduction in bureaucracy. NAPs operate under complex bilateral peering arrangements that must be negotiated among all the parties. These are necessary for two reasons. First, the participants in the NAP have a strong interest in restricting who can be a full participant. This has both economic and technical benefits. The harder you make it to be a first-tier ISP, the fewer there will be, and the bigger the rewards for those that make the grade. Second, not everyone connecting to the NAP is on the same level. Some smaller ISPs who have only a single connection must also have an arrangement--called a transit agreement--with a first-tier provider to ship packets to other NAPs.

With a mini-NAP, there's no need for such a complicated policy structure. By definition, everyone who comes to the mini-NAP is a peer to everyone else. There's no benefit in artificially limiting participation. Also, because every member connected to it has a link to the outside world, there is no need to negotiate transit agreements.

My lunch pal who owns an ISP agreed. He contacted all of the other ISPs in Tucson and, over the next few months, they set up The Tucson Interconnect. It took a couple of rounds to get it right, but in the end, the simplest structure won out: Come to the NAP, pay your fair share, and you're in. It's turned out to be a success, not only for the improved connectivity and reduced load on the big NAPs, but also because we now have stronger relationships in our Net community. Since we had our lunchtime chat, I discovered that there are other mini-NAPs popping up around the world. The benefits are so great and the costs so low that even in cities that have only two service providers, a private link makes sense.

Rally for a new route
So why am I telling you all this? Grassroots pressure, of course. Once you get a mini-NAP set up, it's a breeze to operate. However, most ISPs today are so busy that ventures like a creating a mini-NAP can get pushed off indefinitely. Now that you know about the benefits of mini-NAPs--to you and the Internet as a whole--help persuade ISPs to reprioritize things so a similar project gets off the ground in your town.

If you're using a local ISP, ask customer support whether it participates in a local NAP, sometimes called a regional exchange point (REP) or regional access point (RAP). If not, suggest that it join one or set one up. If your company has a dedicated T1 (or faster) line to a nonlocal ISP, see if there's a mini-NAP in your town that you can join. Joel Snyder is a senior partner at Opus One, an Internet consulting company in Tucson, Ariz.

---

Reprinted from Internet World magazine Vol. 8 No. 1 (c) 1997 Mecklermedia Corporation. All rights reserved.