The continuing saga of Internet connectivity is chipping away at a well-intentioned vestige designed to keep accessibility to the Net squarely in the public domain. Known as public peering, it is a means for Internet service providers to send and receive traffic destined for one another's networks, a fairly innocuous scheme in the early days of the Internet.

The intent was to interconnect at neutral network access points (NAPs). For example, ISP A would have access to ISP B's network, and vice versa, in an environment that treated them equally.

Sounds simple enough. Now, tip the balance so that ISP A is a Tier 1 carrier that is 100 times larger than ISP B. Is the Tier 1 provider still interested in exchanging traffic for free with the smaller player, or does it prefer the traffic of larger ISPs that are similar in size and clout?

On the other hand, if ISP B's bread and butter is enabling his customers to reach other Internet users who could be attached to any one of numerous backbone networks, should his venture suffer? Where does the balance lie?

The dilemma is reminiscent of the struggle to find a level playing field for voice traffic. This scrimmage, however, is waged far beyond the reach of federal control. And it remains to be seen whether the industry can self-regulate and if public peering will survive.

CYBER VISION

Six years ago, the National Science Foundation envisioned how the Internet could survive unfettered with legislative intervention. If the Net was to flourish as a product of the people instead of a governmental tool, the NSF reasoned, it would have to be privately operated. To that end, the NSF helped to create NAPs. It lured service providers to connect to these access points by making participation mandatory for ISPs signing government contracts.

In those early days of connectivity, that carrot was attractive enough to entice ISPs to the original NAPs operated by Ameritech, Pac Bell and Sprint. MCI WorldCom predecessor MFS subsequently opened metropolitan area exchanges - MAE East in Fairfax, Va., and MAE West in San Jose, Calif. - providing public peering anchors on both coasts. In 1996, the NSF declared that the original intent of public peering had been successful, and the foundation pulled its financial support. By that time, the Internet boon was in full swing, and the NSF's departure had little fiscal consequence.

But as the Internet became the darling of both enterprise and residential surfers, the NAPs and MAEs struggled with billowing traffic levels. At one point, the MAEs were home to 95% of all peering. Congestion reared its ugly head, creating latency and packet loss, which turned many ISPs away from public peering and toward private alternatives. With private peering, traffic typically is still exchanged for free, but rather than connecting at a public exchange, two ISPs instead connect their routers at other points in the network.

"The public vs. private peering aspect is essentially one of quality of service and control of connectivity," says Cole Libby, director of network engineering for PSINet, an Internet backbone provider based in Herndon, Va. "[In] the public peering model, you connect to a neutral party and then arrange to have peering connections with other ISPs at that neutral site. Private peering - with direct bilateral interconnections between two providers that are concerned with exchanging traffic and managing the interconnection - is the most reliable way to provide connectivity to the Internet."

A widening gulf soon developed between public and private peering. The largest players minimized their reliance on public peering points and today exchange the majority of their traffic through private peering. But private peering arrangements can be hard for smaller ISPs to come by.

"It used to be that ISPs would peer pretty much with anybody who asked them to, and that created an Internet that was somewhat democratic," says Tom Nolle, president of CIMI Corp., which tracks the telecom industry. "The problem is that the larger players don't gain any business benefit from peering."

Instead, he says, the bigger ISPs "began to charge for transit, the right to move traffic across their networks to reach customers [on other networks]."

Some charge that the dominant backbone providers - UUNet, Sprint and Cable & Wireless - have used peering to retain what amounts to an oligopoly. Combined, those three providers control approximately 75% of backbone traffic.

"It's a boon to them," says Michael Gaddis, chief technology officer for Internet backbone provider Savvis, who recently left to help start up another as-yet-unnamed company. If the three dominant Internet providers decline to peer with another ISP, they often pick up that ISP's transit business - a significant revenue source.

PSINet, which claims to be the largest backbone provider not owned by a telco, has challenged the dominant ISPs' practices while at the same time benefiting from them. It will peer with any ISP - but because many smaller ISPs cannot obtain peering with the Big Three, they often end up buying transit from PSINet to reach end users attached to one of the Big Three backbones (see sidebar on page 15).

THE O'HARE OF THE INTERNET

Recognizing the rise of private peering and transit, operators of some of the public exchange points are seeking to make those points more attractive. One approach is to convert from fiber distributed data interface (FDDI) within the exchange, which has a maximum throughput of 100 Mb/s, to ATM or other high-speed protocols.

Ameritech enjoys a unique distinction among the purveyors of public exchange points as the first to use ATM from the ground up. One of its primary benefits is the scalability it affords.

"ATM has allowed us to scale the individual port speed, so at the NAP, we can support a DS-1, DS-3, OC-3 and OC-12, up from 1.5 Mb to 622 Mb," says Andrew Schmidt, Ameritech's senior product manager. "That's tremendous scalability in the port speed."

Like its aeronautical neighbor, the Chicago NAP is one of the busiest hubs in the United States, adding roughly one new ISP per week. But because its customers use Web-based permanent virtual circuits (PVCs) for provisioning, it works best when it works invisibly, according to Schmidt.

"We really don't want to be involved in the customers' day-to-day operations, and we aren't - we're transparent," he says. "In Chicago when a customer connects on an ATM circuit, by default we create a full mesh of [PVCs] between them and all the other sites. We have a simple number scheme that is posted on the Web page to show how to allocate virtual circuits."

The advantage to this, he explains, is that when an ISP connects, it immediately can peer with other providers. They don't need to ask Ameritech to intervene.

"We try to make the ATM exchange port look like a big buffering Ethernet that works from the moment you connect," Schmidt says. "The last thing we want to see in an exchange point is traffic being lost. We've gone out of our way to use this ATM switch with extremely large buffers."

Ameritech was well ahead of the public peering curve when it deployed ATM and PVC. MCI WorldCom followed suit last year byadding a PVC product called PeerMaker and, in May, rolling out an ATM for MAE East, MAE West and its new MAE Central in Dallas. It was the first public exchange to light OC-12 for connection.

These upgrades, according to MCI WorldCom, answer the ISPs' call for increased bandwidth, consistent reliability and more traffic control. But the ATM switch currently does not connect to the MAE West's FDDI switch, which limits peering options for new backbone providers. Such limitations lead to speculation that MCI WorldCom, which owns UUNet, might have an ulterior motive - to encourage ISPs to move to private peering or transit relationships.

Caleb Williford, MCI WorldCom product manager for MAE services and Internet access, denies such allegations.

"Both the private peering and the public peering points have grown in parallel," says Williford. "Quite simply, that's the bandwidth explosion we continue to see. I don't think the industry could head toward an all-private or an all-public peering environment. From a redundancy standpoint, I think it's a great idea that people augment their networks through [both types of] peering."

WHERE THERE'S A WILL...

The fray over peering - or for some, the lack of it - is spurring new ventures that fill the void in Internet connectivity.

Perhaps one of the best of the breed is Equinix, the brainchild of Cisco Systems, Microsoft and Benchmark Capital. Neutrality is the centerpiece of Equinix's grand design to launch a global network of some 70 Internet business exchanges, the first of which is slated to open this summer in Washington D.C. From these local connectivity centers, Equinix customers can make private connections free of the congestion that slows traditional peering.

"The original exchange points were built around a central switching concept," says Jay Adelson, Equinix's chief technology officer. "But any central switching concept is limited, and eventually that switching fabric can saturate. At the rate the Internet is growing - approximately 30% a year - you can't keep technology totally ahead of that scalability. It's important that you create a private interconnect model that bypasses the central switching concept."

AboveNet, the Internet backbone provider that recently purchased the Palo Alto Internet Exchange, also plans to allow both public and private peering within the exchange (see sidebar on page 16).

Other Internet backbone providers also are focusing their strategies on solving problems associated with traffic exchange between ISPs.

The Internet connections of some large business customers currently operate at only 20% of capacity, says Gaddis, formerly of Savvis. He believes the opportunity for improving that throughput lies in understanding how traffic is exchanged between Internet providers.

"When you connect your second DS-3 you are, in a sense, a backbone company," says Gaddis. "You have to engineer your network with a knowledge of how the backbone is constructed."

Border gateway protocol (BGP), the routing protocol commonly used between Internet peers, is stupid, Gaddis says. "[It] requires that you know more about the backbone network than BGP will tell you."

Savvis' approach is to buy transit from multiple backbone providers and to send each backbone operator only the traffic that is destined for end users attached to that backbone, thereby improving throughput.

Citing estimates that $800 billion in business-to-business transactions will be conducted over the Internet within a couple of years, Gaddis says, "The money will flow into providers that can solve this problem."

Referring to Savvis' acquisition by financial information firm Bridge Information Systems, he says, "If we can solve the problem for the financial industry, we can replicate that for other industries."

Level 3, the fiber network operator that also has built an Internet backbone network, is taking a different approach toward addressing traffic exchange problems. The company hopes to sell user pairs on the concept of using Level 3 as their provider at both ends of the connection, thereby avoiding inter-network traffic exchange altogether.

"We're trying to get away from the current peering and transit model," says Tom Sweeney, senior vice president of strategic alliances for Level 3. "It doesn't do the right thing [with regard to] quality of service."

Level 3's offering, dubbed CrossRoad, also allows customers to avoid transit charges on connections to other Level 3 customers and has been well received by content providers, Sweeney says.

ONLY THE STRONG WILL SURVIVE

Ask telecom veterans whether public peering will survive and you're likely to get a broad array of responses from one extreme to the other.

"The easy answer, and the obvious one, is yes," says Ameritech's Schmidt.

"Absolutely," predicts Michael Brookins, manager of system engineering for AboveNet. "The only way for the Internet to survive is with public peering."

CIMI's Nolle believes one of two things will happen. "One, we're going to build what I call a shadow Internet, meaning the public carriers that have statutory interconnect requirements are going to interconnect their underlying infrastructure in the historic regulated way," he says. This will become the foundation for a new public IP network that will serve a lot of the purposes that today's Internet serves, he explains. It will look exactly like the Internet, but it will be dominated by facility-based carriers because they can at least agree on interconnection.

The other possibility is that the industry will see some regulation in the Internet space, Nolle explains. "If one of those two things doesn't happen, the Internet will crash eventually."

PSINet is unique in several ways, but perhaps the most apparent difference between it and other backbone providers is that PSINet is the only major player not connected to a telco. Because of this, it isn't surprising that the company's strategy has one major theme: acquisition.

"Our primary strategy for growing our backbone is through the acquisition of capacity," says Bill Johnson, PSINet's senior director for U.S. network implementation. "We do this to reduce costs and to be able to offer more bandwidth flexibility."

The company has made large purchases of both lit and dark fiber. An agreement with IXC gave PSINet 10,000 route-miles with OC-48 capacity - 3300 miles of which are already in use. It also continues to purchase dark fiber from Metromedia. "With Metromedia and IXC, we're following their builds," Johnson says. "We don't dig trenches, and we don't lay conduit."

The other prong of the acquisition strategy lies in purchasing ISPs. "We just bought one called TIAC, and last year we bought Ionet. Internationally we've bought many ISPs. That's our main strategy for international growth," Johnson says.

PSINet, he explains, is the only non-telco provider that owns capacity that can go around the world. This non-telco status is a main differentiator for the company, Johnson says. It gives the company some flexibility that telco-related ISPs may not have. "It's an advantage to go into the marketplace and focus entirely on IP."

But the company has other differentiators. Unique attributes include PSINet's strong focus on business-to-business Internet within the U.S., as well as the number of points of presence it owns - more than 250 in the U.S. and 535 globally. "We're in all the NBA and NFL cities, and we're trying to expand to Tiers 3 and 4 through acquisitions or by extending our fiber out," Johnson says. "We're not done with just 10,000 route-miles - we're looking for both dark and lit fiber in the U.S. and Canada."

Operating data centers is also part of PSINet's strategy, Johnson says. Data centers enable Web-centric companies, such as portal operators, to move their content to carrier-grade facilities located as close to the Internet backbone as possible. Often content can be replicated from one data center to another, providing redundancy and minimizing the distance information must travel to reach end users.

This is a strong reason for continuing to buy more capacity. "As more data centers are introduced, it will change traffic flows. By owning more capacity we can accommodate these changes in traffic. "Within the backbone market, PSINet has distinguished itself by offering free peering agreements with regional ISPs. This strategy allows the company to collect transit fees from customers for crossing its backbone to connect to the Big Three networks. But Johnson doesn't see those major players as a dominating force. "We think that by focusing on a business-to-business strategy, we have a strong presence in the U.S.," he says. "We are not an oh-by-the-way company."

AboveNet is a company that's hard to pigeonhole. It has a nationwide fiber network and provides Internet connectivity to that network for ISPs. It has 257 peering agreements and owns the Palo Alto network access point. A big part of its business, however, is providing co-location for what it refers to as Internet-centric companies - Web developers, Web hosters and those that provide video applications, e-commerce and other high-bandwidth applications. It also manages the services of such providers and offers peering within the co-location facility.

The other thing about AboveNet - it was purchased in June by Metromedia Fiber Network.

"We will keep the same AboveNet name and the same AboveNet business policies," says Van Jepson, AboveNet's vice president of marketing. "The only difference is that AboveNet stock will be converted to Metromedia stock."

The major benefit of the agreement will be that Metromedia's network will enable AboveNet to reach speeds of more than a terabit per second. "This will decrease the cost of business and the time to market for many of our clients," Jepson says.

Connectivity for ISPs, he explains, is a key part of AboveNet's business - the company's main reason for purchasing the Palo Alto NAP, in fact. He says AboveNet should truly be seen as the "Net above the Net" because it is a network of peering relationships that runs over the company's fiber. "We want to be a place where ISPs and content providers come together."

Important to the business is its technology, Jepson adds. One kind of technology it uses is its multi-router traffic grapher, a management system that "gives a potential client the ability to make sure everything is up and running," he says. It allows for customized billing and offers the ability to forecast network growth. "This is important because we can monitor where we peer with others, and we can see aggregation points and predict the growth rates of those points," he says. Customers that use this technology include DoubleClick, Netscape and MSN Hotmail.

Another important technology for AboveNet is asymmetrical allocation of packets, which uses a protocol that automatically routes packets to the fastest path in its network. "It's a very important tool," Jepson says. "It minimizes packet loss and delay, and helps us offer a more guaranteed service."

The company employs an open business model, meaning that it will not compete with clients or partners. Therefore, even obvious competitors are looked on as potential partners, Jepson says. "We focus on our network."

The Palo Alto exchange is a public peering site, but AboveNet will allow private peering within the site. The company also runs dark fiber from the Palo Alto exchange to MAE East and MAE West, fueling speculation that it might be seeking a means of enabling an ISP connected to Palo Alto to peer with ISPs connected to the MAEs. That would be no small feat, considering that MAE West has not accepted any new ISPs for more than two years.

Jepson says the Palo Alto exchange will expand AboveNet's "openness" both nationally and globally. "It's the best model out there, and we want it to proliferate," Jepson says. "This will create a better experience for ISPs and for those who put content on our network."