Peering and Fearing: ISP Interconnection and Regulatory Issues

By Kenneth Neil Cukier *

All, all of a piece throughout:

(To DIANA) Thy chase had a beast in view;

(To MARS) Thy wars brought nothing about;

(To VENUS) Thy lovers were all untrue.

'Tis well that an old age is out, And time to begin a new.

—John Dryden, The Secular Masque.

* * *

Introduction

Is some form of government- or industry-led regulation needed among Internet service provider interconnection agreements to ensure the ISP industry remains open, fair and competitive? It's a tough question, since regulating the Net to better ensure a competitive environment might have the reverse effect and burden the dynamic, nascent sector.

At the heart of all networks is interconnection—the ability of one entity to connect to other entities. For the "network of networks," the Internet, such interconnection isn't only vital for users, it represents the very essence for which the Internet stands. The Net is built upon open standards, with as little central control as is feasible. The result is any network via any means of communications can access users of another network in an easy and inexpensive way. And throughout the Net's history, these interconnections have generally occurred on a settlement-free basis, known as peering. The Internet's epic success speaks for itself.

But, like all epic tales, there's a hitch. Without interconnections, many of the benefits the Net has so far realized as an open system could be quashed. Users on one ISP's network would not be able to reach users on another ISP's network. Or, if the terms are commercially discriminatory, then the price of Internet access could artificially skyrocket and new market entrants be disadvantaged.

In general terms, interconnection is a historically tangly issue since the development of network systems like railroads or the telephone network has often been accompanied by the market dominance of a handful of players who abuse their position for commercial gain by restricting the interconnection of rivals.1 In the United States, the justification for telecommunications regulations that focused on interconnection was due to the legacy of the Bell system, which evolved into a quasi-monopoly for what was considered a public good.2 The issue is still with us: the ability of public officials to mandate and regulate interconnection agreements among telecommunication carriers was re-affirmed in the Telecommunications Act of 1996.

As regards the Internet specifically, interconnection became a controversial topic once commercial backbone provision began in the early 1990s. Yet the matter reached a crescendo between 1996 and 1998 when large backbone ISPs began radically changing their interconnection terms—and in so doing, critics charged, changed the original personality of the Internet from enthusiastic engineering to clawing commercialism. Today, large networks aim to transition peers into a supplier-customer relationship.

This chapter attempts to denote the changes that have taken place in the industry and identify how certain aspects inherent to the Internet lead to conflicts over peering. Finally, after analyzing the potential for market abuse by dominant networks, a framework to address the problem is offered along with an examination of how it might be put into place. Although the amount of information regarding ISP interconnection has increased recently in academic circles as well as the press, very few works specifically address the matter in the context of public policy. Yet industry developments have progressed so swiftly that a decision whether to regulate ISP interconnection is squarely on the agenda of European and United States regulators at the time of this writing, in early 1998.

Interconnection is fundamental to the open Internet. In Request for Comments (RFC) 1958, an official document of the Internet Engineering Task Force, the IETF sharply states: "the community believes that the goal is connectivity."3 However concerns have been raised by small Internet service providers that the fabric of the Internet is fraying due to new policies by large, so-called "Tier-1" networks to peer only with the largest of siblings. The Tier-1 backbone ISPs, however, counter that the cost of infrastructure investment to sustain growth necessitates they recoup this cost via settlement fees from the smaller providers, unless for some reason it is more beneficial to exchange traffic on a settlement-free basis. Peering is generally judged worthwhile when the traffic exchanged between networks is roughly equal, or there is a commercial and technical benefit, such as improved quality of service. A decision not to peer is usually made if it is technically cumbersome to do, or there is a wide traffic imbalance between the networks. Additionally, the matter is made more complex since the Tier-1 ISPs themselves are not clear on which way a settlement should go if there is a traffic imbalance: Should the sender pay, or the receiver?4

Farnon and Huddle have argued that settlement-based interconnections achieves "efficiency" because it "is technically workable, fairly compensates all providers, and promotes interconnection among networks."5 However, the debate framed in this way addresses only half the issue: it ignores the impact of a settlement-driven business model that may give rise to a concentration of market power. One can accept their approach that downstream ISPs should pay for upstream transit, and still raise new questions their work did not address, notably the criteria backbone ISPs use for peering, and how that criteria is made known to the marketplace. The answers affect those similar-sized networks that also sell transit to downstream ISPs, and the ISPs that are slightly smaller by some measure, which, on the borderline of the group, keenly eye breaking into the more elite circle of large backbone peers.

Bailey follows on ground tread in the telecom world with its long-established principles of network evolution to conclude: "As the Internet matures into an infrastructure with dominant market players, older interconnection agreements may no longer be consistent with a business' competitive strategy."6 Bailey's work analyses the different economic incentives associated with different types of Internet interconnection arrangements, and is not intended to consider regulatory issues.

Other chapters in this volume discuss ISP interconnection matters. Friedman and Mills-Scofield, on a purely economic perspective, examine the optimal settlements pricing strategies for ISPs. Gideon widens the viewpoint on interconnection by seeking "first principles" for interconnection pricing that apply to both the ISP and carrier networks. The approach taken here is more reportage—which perhaps befits a sector marked by rapid change, and the author's background as a journalist. That the points presented may merely end up a matter of historical record due to the Net's dynamic pace is a burden all writers on Internet themes share.

Lastly, the chapter discusses peering as it pertains to the United States. Certainly the issue has direct international implications in areas such as foreign-owned networks interconnecting with US-based ISPs, settlement charges based on the international accounting rate regime, and the trend that foreign carriers pay the full cost of circuits to the United States. On a policy level, it is probably impossible to separate the international dimension of the issue from a national one. However conceptually, the paper concentrates on the United States' experience since it is where the Internet sector is most evolved, and it serves as a petri dish to consider the broader issues that may cross-apply internationally.

A Definition of Peering

Although the term "peering" is used frequently, it rarely has a uniform meaning. For example, there is no set definition for it in the IETF's official documents, the Request For Comments series. RFC 1983, which provides definitions to important Internet-related terms, has no entry for "peering" or for "interconnection", "settlement" and "sender-keeps-all."7 And the term is used among network operators to mean any type of network interconnection, such as a "BGP peer,"8 which lacks the deeper significance of whether the traffic exchange is based on a "sender-keeps-all" (SKA) model or fee-based settlement model. The term is also at times used to include transit to third-party networks.

Geoff Huston, an Internet engineer at Telstra Corp. in Australia, refers to three peering models in an early draft paper on settlement issues.9 The second model he describes most closely corresponds with the common understanding of peering. It is: "Sender Keep All (SKA) where each [network] invoices their originating clients user for the end to end services, but no financial settlement is made across the bilateral peering structure."

An important distinction is that peering does not mean transit to a third party. Rather, it is the exchange of one ISP's customer's traffic onto the network of another ISP's customer, to whom it will be delivered. Also significant is that Huston's definition does not account for multilateral peering arrangements. This simplifies things.

Farnon and Huddle note that peering historically developed from the interconnection of similar-sized networks. They defer to Gerald W. Brock's two conditions for the SKA model to be viable: That traffic is balanced between networks and that the cost of terminating the traffic is low compared to the cost of metering it.10

For our purposes here, ISP peering can be understood as: An interconnection of two public networks that provide connectivity to hosts whose routes are advertised on the global Internet, on a settlement-free basis that allows customers of one network to exchange traffic to customers directly on the second ISPs' network. This definition is a fuller description than simply saying "SKA interconnection." Generally, the "peers" will be of roughly the same size—recalling the traditional sense of the word. The notion of similar size might be determined by number of customers, backbone capacity and traffic volume, or any other factor established by the networks that is used as a basis to exchange traffic freely.

Developments in Peering

Recent studies and Internet industry developments have brought the peering issue important attention, which gives context for the present discussion.

In the fall of 1996, at the conference "Coordination and Administration of the Internet" sponsored by the Information Infrastructure Project at Harvard University's Kennedy School of Government, certain papers briefly alluded to whether peering issues posed a need for government intervention.11 Farnon and Huddle, for instance, note: "If the trend sparked by the SKA settlement model continue and thus discourage networks from interconnecting, then the government might have reason to mandate interconnection among networks."

In October 1996 two large ISPs, Digex Inc. and AGIS, cut off their peering connections for over a week due to a dispute. Foster notes the significant impact on the Internet industry and consumers: "AGIS customers were not able to access Web sites that were on the Digex network including the Security and Exchange Commission Web site."12

During the winter of 1997, UUNet Technologies Inc., MCI Communications Corp. and BBN Corp., key members of the Commercial Internet Exchange, left the CIX router, which was the first commercial interconnection point. At the February 1997 meeting of NANOG in San Francisco, Randy Bush, a network engineer at Verio Inc. noted that peering was a looming controversy due to recent reductions in peering relationships, lack of publicly stated criteria, and abuse by downstream ISPs that set "default" routes to the backbone ISPs, thus requiring the larger ISPs to carry the small ISP's traffic rather than the small ISP's direct upstream provider.13

Between March and May 1997, UUNet told around 15 ISPs that their peering arrangements would be terminated within a few months, and that new, bilateral transit agreements must be struck. The move essentially transformed peers into customers. In one instance, the public shake-up resulted in the dismissal of a network executive, David Holub, then-president of Whole Earth Networks Inc. in San Francisco. Holub soon thereafter posted a public message on the North American Network Operators Group (NANOG) e-mail discussion list arguing that such peering cuts were anti-competitive and possibly illegal. He further stated that the government—either on a federal level, or more likely state level—had a role to play to protect the public interest by seeing that peering criteria is made public and not used in a discriminatory way.14 A fuller discussion of Holub's argument appears below.

Throughout the year, large ISPs transitioned their peering arrangements away from so-called "public media" of large exchange points in favor of setting up direct interconnections with other networks. The engineering rational is that it allows networks better oversight of the technical aspects of interconnection.

During this time, although backbone providers eliminated many peering arrangements, peering grew more prevalent among smaller networks, and was boosted by the creation of more local network access points.15 In August, one backbone ISP, PSINet Inc., said it would "peer" with any ISP so long as the other network met certain publicly stated requirements. Additionally, a national network built up over the course of the year by merging with regional networks, called Verio Inc., became a major backbone provider. It is assumed that Verio was able to slip into the Tier-1 ISP circle. However, due to non-disclosure agreements, this cannot be formally ascertained.

Alongside these events, numerous Internet backbone providers were bought by larger ISPs. Specifically, the consolidation decreased the numbers of ISP backbones, and widened the margin between Tier-1 providers and other ISPs.16 Of special note: GTE Corp. bought BBN Corp. in May. UUNet announced their intention to buy ANS Communications Inc. and CompuServe Network Services in September. In November UUNet's parent company WorldCom Inc. and MCI agreed to a merger. GTE bought Genuity Inc., also in November. (Parallel to the ISP consolidation was a trend for backbone ISPs to become aligned with facilities-based telecommunications carriers. Some examples include the Internet's first commercial ISPs, such as PSINet Inc., Netcom On-Line Communication Services, Inc. and CERFnet Inc. While this has not limited the raft of possible competitors in the backbone network sector, it does underscore the importance of facilities-based Internet provision in order to remain competitive.)

The impact of the consolidation means that fewer players control more of the traffic on the Internet backbone. Measuring ISP backbone traffic is difficult, as is defining what parameters one wishes to measure. However, analysts and researchers say that by November 1997 the US's four largest networks (UUNet, MCI, BBN, and Sprint) controlled between eighty-five percent and ninety-five percent of the total backbone traffic.17 Foster in late 1996 estimated "ninety-five percent of all Internet traffic passes through these [top] five [U.S. Internet backbone] networks."18

In terms of politics and regulation, the U.S. follows a market-led approach. A White House report issued in July called "A Framework for Global Electronic Commerce" seeks a non-regulatory, industry-led system to govern (as opposed to regulate) the Internet from an internationally-minded basis.19 A bill has also been introduced into Congress that would prohibit the FCC from regulating the Internet.20

Additionally, Billy Tauzin, the Republican representative from Louisiana and chairman of the House subcommittee on telecommunications, said before a congressional hearing in October (speaking about WorldCom's bid for MCI) that even if sixty percent of U.S. Internet traffic is controlled by one player, he still believes there is healthy competition since many rivals exist or are forming.

Yet the Federal Communications Commission's public stance is ambiguous. Throughout his tenure, former Chairman Reed Hundt used every opportunity to indicate the Commission opposed Internet regulations, since the Internet represents a new medium characterized by enormous growth and rapidly changing technological innovations. Such regulations, he argued, might either address the wrong problem, not be effective, and could possibly harm Internet development. With four new commissioners appointed in autumn 1997, whether this perspective continues to hold sway is still to be seen.

In a FCC working paper21 issued by the Office of Plans and Policy in March 1997, the Commission noted the Internet falls under its remit, regardless if it decides to impose regulations. It sets up a test of where government's role might be appropriate: "... government should create a truly level playing field, where competition is maximized and regulation minimized."22

The word "peering" only appears once in the FCC report, to explain that backbone ISPs must interconnect to exchange traffic. Where the report does discuss interconnection matters, it does so in terms of whether dial-up ISPs should pay access charges to local exchange carriers, which ISPs, as enhanced service providers, are currently exempt from doing.23

The FCC report explains this is not analogous to the Internet backbone sector, and thus the question of Internet backbone regulation is not treated, stating:

"... prices that carriers charge for use of local exchange facilities are regulated, while those that Internet backbone providers charge are not. This regulatory distinction is based on the reality that today there is generally only one LEC that an ISP can use in a given area, while there are many competing Internet backbone providers."24

However, the questions that arise are whether market dynamics changed over the past year and the consolidation mean less competition, which encourages anti-competitive behavior? If so, what is the best way to address the matter in the least restrictive way? Before treating these questions, the impact of the recent events on the marketplace for peering needs to be considered.

Current Market Characteristics for Peering

The events have come in tandem with a radical change in the way peering contracts are established. This was not a surprise. Some network engineers say they saw the trend emerging in the early 1990s. They point to the emergence of the CIX in 1991 as an early warning—and industry-driven solution—to the matter. CIX was a cooperative, multilateral peering point. Researchers and industry observers also predicted higher priced Internet access, massive consolidation among networks, new settlement models and the affect integrated services may have on prices for Internet connectivity.25

The recent events, together with a state of flux for Internet pricing models, sets the backdrop for the current peering situation between Tier-1 ISPs en masse, and between Tier-1 and Tier-2 ISPs.

Over the past year, the following trends stand out:

* The number of Tier-1 ISPs has shrunk due to consolidation.

* The margin between Tier-1 and Tier-2 ISPs has grown wider (measured in amount of infrastructure, number of customers, and backbone speed).

* Tier-1 ISPs have decreased the number of ISPs they peer with.

* The terms of peering are not made public by Tier-1 ISPs.26

* The ISPs that peer with Tier-1 ISPs are not publicly disclosed.

* ISPs who peer with or interconnect in another way with a Tier-1 ISPs must first sign a non-disclosure agreement.

* Interconnection and peering arrangements can in cases be discontinued with little notice on the part of the larger network.

* Tier-1 providers (and other large networks) meet regularly and privately to discuss engineering issues.

The traditional terms for peering, broadly noted, are:

* Backbone capacity: Can the peer adequately handle the traffic flow?

* Presence at peering points: Can the peer transfer or accept its customer's traffic onto a network at many points?

* Symmetry: Is the volume of traffic exchanged roughly equal?

* Quality: Does a peering arrangement allow for better quality of service between networks, or provide better access to a sought after site?

* Operations: Does the peer have a satisfactory network technically? (This generally includes a network operations center open 24 hours a day, seven days a week.) Does the peer use the same form of routing protocols?

However, the technical, economic and contractual models used for peering arrangements remain in a state of flux, and ISP engineers and executives blame the problem on the way the Internet functions. A few characteristics of the marketplace are identified as exacerbating the peering issue.

Arbitrage

The lack of an industry-wide approach to determine a peer affects the overall dynamics of the Internet sector by creating a form of arbitrage that is played behind the scenes by the different ISPs of different sizes when negotiating new interconnection contracts. Backbone ISPs that provide connectivity to smaller ISPs yet must also interconnect with larger ISPs act like foreign exchange arbiters: they seek to extract revenue in both directions.

Consider the case of two backbone ISPs, in a case where ISP-A is larger in terms of national presence and customers, but receives a traffic imbalance when peering with ISP-B who sends more traffic to A. ISP-A requests a settlement fee from B (and perhaps threatens to cut off peering to do so), because B is sending more traffic over A's backbone. The flow is unequal, says A, and A has to expend more network resources to carry B's traffic. B complains that a settlement fee is unjustified: A's customers, who are requesting the content, are the ones who should pay A more for the service of using A's backbone to transport it.

ISP-B, however, would be in a position to turn ISP-A's argument around. B says that A's customers want to access B's content, and to cut off the interconnection would deny A's own customers a proven popular service. The traffic imbalance, according to B, suggests that A is in fact beholden to B and not the other way around (as A had first suggested!). Although ISP-A, the larger network, would deny more customers to B than ISP-B would to A, a dis-interconnection would result in a sort of "mutually assured destruction." The positive network externalities would be quashed for both ISPs. (There may be a calculation that could measure which ISP would "feel more pain" by the dis-interconnect, but I leave this to the mathematicians.)

Moreover, feeling empowered, B takes the argument a step further: ISP-A's customers place a strain on B's bandwidth resources by their demand for B's content. As a result, ISP-B requests the similar conditions on A that A formerly placed on B, namely: "you require me to expend greater network resources, so you should pay me to upgrade my connection to you or I will cut off our peering." In this instance, now B is claiming a settlement (justified by the need for better infrastructure to handle A's customer's requests) from A. The smaller network B has fully turned the tables on the dominant ISP-A.

So: Who pays whom? For the moment, there is no set industry approach. Comically, what happens in the marketplace is a single ISP acts like A with one peer and then acts like B with another, to fit a given situation. With a lack of public disclosure, this continues and all ISPs seek to gain in the arbitrage.

The Hot Potato

The scenario between ISP-A and ISP-B is especially relevant in regards to ISPs that primarily host Web sites. Such Internet value-added services are a highly profitable area since basic Internet access resembles a commodity, and many ISPs now specialize in hosting. Indeed, these ISP distinctions are a sign of the sector's new maturity. In instances where one ISP exclusively hosts sites, the traffic flow will by nature be imbalanced. The hosting ISP may have a good connection at one or more public interconnection points, but the flow is one-way. Other ISPs' customer's are sending a few bits of data to the host, to request a Web page. The host then sends off a large amount of traffic to the user.

Because of "shortest exist routing," the standard means of exchanging traffic among networks, the data goes onto the receiver's network at the earliest point. All networks have an interest in passing off the data, like a hot potato, to the other network at the earliest point, since it saves the sender-ISP valuable bandwidth. Besides, the argument goes, the receiver has paid his or her network provider for the Internet service and not the hosting service.

In terms of peering relationships, shortest exist routing means that a backbone ISP at times finds it must haul traffic across the country on its own network, rather than the host site's network carry the data to the closest point to the customer, where the customers' network delivers it on the portion of their infrastructure closest to the termination point. When networks were all roughly the same size, the game of "hot potato" was treated as a simple reality of Internet provision and accepted since the balance was considered equal, or at least not extremely different (or the network externalities from interconnection so convincing) that the traffic was not worth metering.

Now, however, as ISP niches become more distinct, the benefit of shortest exist routing is called into question, since it makes for an inherent traffic imbalance between networks that supply connectivity to content sites (and are net exporters of data) and those networks with many subscribers that request the data (and which flows greater over their own provider's network).

What is the Internet?

Implicit in the notion of an ISP is the expectation that it provides connectivity to the Internet. But this begs the question: "What is the Internet?"27 This is not as infantile a question as it may at first seem. It is very relevant to the peering debate, as well as whether regulators may carve a role to play in Net matters.

Is the Internet all hosts with Internet Protocol numbers across all networks everywhere, or is it some percentage?28 The issue dovetails with interconnection among Tier-1 ISPs and Tier-2 ISPs. The latter have a national network, but are not so large as to command a proportion of the Internet that if a Tier-1 ISP cut off their routes, the Tier-1 ISP would lose connectivity to the global Internet.

Instead, the opposite is true: The Tier-2 ISP is beholden to the larger network, since the network externalities associated with interconnecting with the Tier-1 are greater for customers of the Tier-2 ISP than is it for the Tier-1 ISP's customers to interconnect with a network of fewer customers. This marks a profound imbalance in the power the two networks wield—and how they come together to negotiate interconnection. Importantly, if the larger network cuts off the smaller, the larger still has a sizable reach of the Internet. The smaller is worse off. But can it be said that the smaller network is "no longer a part of the Internet?" If the large network has a dominant market position, say a majority of all backbone traffic, than perhaps the smaller network is no longer selling "Internet" connectivity if it is excluded from reaching customers of the larger ISP. The smaller network in that case is forced be a customer of the dominant network or another upstream provider to connect their customers with the customers of the dominant network. (The issue is equally precarious, yet not as drastic, for the larger ISP. Does it still sell "Internet access" if they are unable to interconnect with certain networks, even though it makes up a very minor fraction of the "Internet"?)

The voice telecommunications network is founded on the principle of universal connectivity. A handset, a subscription, and a number is understood to mean the customer can reach all other numbers and can also be reached.

The Internet, however, lacks a specific definition, and it is uncertain whether the telephony model applies to it. The original purpose of the Internet was nothing more grandiose than to share remote computer resources among several dozen sites. It was a data communication protocol over a wide area that employed packets and was capable of withstanding a central point of failure. Later, the idea of global connectivity was fostered, as ever-increasing numbers of individual networks connected to the Internet. The early designs of the protocol was intended to make the communications easy for any media of communications to connect, such as wire line, satellite or radio.

Although it is a myth to say global connectivity was an initial Internet principle, today the matter is moot. ISPs sell "Internet access" or "Internet service" without making the distinction whether the provider furnishes connectivity to every IP number ever allocated, such as those allocated to the U.S. military network. In fact ISPs as a principle don't provide universal connectivity. Corporate networks, for instance, can and do use IP numbers for their internal networking needs and the routes are not advertised outside their sub-network to the global Internet. Nevertheless, Internet service provision assumes that the entire commercial Internet is accessible. That is clearly the spirit of RFC 1958. A dis-interconnection by networks poses the question whether either ISP then is truly offering Internet service.

How the Internet is defined will have important implications on whether and in what areas public policy may have a role. Regulators may employ a definition of Internet access to ensure that networks Interconnect on traditional regulatory grounds of consumer protection and labeling. However, engineers do not have set answers to what constitutes Internet service, as testified by the discussion on the subject on the NANOG e-mail list in May after UUNet's decision to de-peer was made known.

These three market characteristics for peering—connectivity arbitrage, shortest-exit routing, and route accessibility—inherently create an environment where market dominance can be abused.

Potential for Market Abuse

Government is constantly torn between whether to regulate to stave off future potentialities or deal only with current realities. However, in the words of one FCC official in a private conversation with this author: "We'd like to know if in three years down the road we need to regulate, was there something we could have done today that might have prevented the situation from happening?"

In that vein, this section is analytical. Instead of demonstrating that (one or more) dominant networks currently exist and abuse their market position regarding network interconnection, it aims to show that the dynamics of the Internet industry is likely to give rise to a situation where one or more networks could abuse its (or their) power.

Driving this are factors inherent to the Internet and in the nature of networks in general. Regarding the Internet sector specifically, the lack of information and alternatives among backbone networks gives a dominant ISPs an unfair advantage. Head to head with another backbone, the larger ISP can require a settlement or threaten to de-peer which might leave the smaller network worse off. Facing regional networks, the larger ISP—or club of the top ISPs—can set the terms of interconnection outside the realm of normal competitive pricing constraints without the natural oversight that comes with a marketplace where essential price information is known among rivals.29 More generally, however, the evolution of networks suggests that abuses are commonplace unless kept in check, either by competitive market forces or by some form of regulation.

Taken together, these lines of analysis form the rational for an open framework for peering that concludes this section.

Perspectives of Large- and Medium-Sized ISPs

The approach to peering matters is a function of the size of the network. As Bailey and others have pointed out, there are economic incentives to chose one form of interconnection agreement over another to optimize efficiency and recover sunk costs of network infrastructure. However, there is a tension between the two camps due to the different value placed on the interconnection. For a backbone network, peering with a regional ISP comes at a cost, since the smaller player will derive more benefit, i.e. the national infrastructure of the larger ISP.30

It is the position of large, Tier-1 ISPs that the decision to drop certain peers is due to a natural winnowing process in the industry that is necessary if the sector is to develop. These ISPs note the process simply re-adjusts the rapport among providers to account for true market position, that of customer and supplier. The revenue garnered from transitioning peers into customers are justified by the need to finance the large network build-outs and increased capacity, from which all networks derive benefit, be they direct downstream ISP customers or peers.

In this perspective, the rearrangement of peering agreements is due to healthy competitive pressure to create greater market efficiency and cost recovery, which makes for a vibrant and innovative sector. Network quality of service is said to improve because the larger network doesn't risk passing off customer traffic to smaller networks that are under-prepared to handle it. Thus customers benefit, as they do from the low cost of access since the networks are considered to operate on economies of scale.

Not driving the decision to de-peer, these networks say, is their dominant market position or any form of collusion. The Tier-1 ISPs state that alternatives exist for networks that have been de-peered by one network to peer with other networks. "If you do your business plan and the only way to succeed is to get all your raw materials for free, then you don't really have a business plan," said the network manager at one of the largest Tier-1 ISPs during an interview.

Indeed, it is not surprising that large network service providers are keen to explain the benefits of buying rather than bartering access. Gerrese, speaking for AT&T-Unisource, argues the advantages to smaller networks of being a customer are: "you get access to all other customers and all peering partners. You don't have to peer yourself [and] you can focus on your own business." He also reveals the disadvantages of peering, noting: "you have to continue to make peering costs" and "you face [the] risk to lose future peering."31

The fear among mid-sized networks that have national infrastructure is that they are not requesting "all raw materials" be free, but the exchange of traffic be free for customer-to-customer traffic. It is not transit they seek, but the customer's routes.

This represents a stark division in the value of interconnections among large networks. Quantitatively, the cost of the interconnection is network resources, i.e. the bandwidth expended to carry traffic from another network onto one's own network for one's own customer. Qualitatively, the value is the connectivity, the interoperability of customers. Large ISPs say they subsidize the smaller networks due to quantitative cost of peering. The smaller, but national, networks maintain they bare a complimentary cost for peering in terms of network resources (they state they have national infrastructure for carrying traffic), but without peering, are hurt more that the larger network in a qualitatively sense. This disparity gives rise to a condition where a dominant network can abuse its market position.

Tier-2 networks say the lack of public peering criteria places a glass ceiling between the two tiers even if they accept the need to operate at a similar scale as the larger networks and actively seek to do so. Non-disclosure agreements can apply in cases where one network shares sensitive business and engineering data with a network to establish the interconnection, but the basic parameters of what constitutes a peer should be made known.

Indeed, these concerns were born out in a recent case where one network told others that peering arrangements would be cut off, and that the smaller networks needed to become a customer of the larger network to access the larger network's customer's routes.

In the spring of 1997, UUNet informed Whole Earth Networks Inc. in San

Francisco (along with about fifteen other ISPs) that its peering status would be discontinued. Whole Earth's president, David Holub, resisted the move in late April, citing abuse of market power by UUNet. In May 1997 he was fired for challenging the decision of Whole Earth's owner, who agreed to UUNet's demand for "paid-peering."

Days later, in an e-mail posting to NANOG, Holub presented a case to show that a minimum of regulation is needed to protect smaller ISPs and their customers.32 Additionally, he indicated that his perspective fit the spirit of the 1996 Telecommunications Act. His solution was for state public utility commissions (PUCs) to regulate Internet peering arrangements, and require disclosure to prevent discriminatory practices. PUCs are allowed to regulate interconnection agreements among interexchange carriers and local exchange carriers under section 251 of the Act in cases when the market players cannot come to an accord.

Houlb treats Internet interconnection as analogous to telecommunications carrier interconnection.33 He believes regulation is justified on the grounds that "universal reachability" of networks is in the public interest and thus requires regulatory oversight to see that the interconnection occurs in fair, non-discriminatory ways, with time limits to comply, arbitration procedures, public disclosure and regulatory approval of terms.

His argument, however, clearly has weaknesses. He never demonstrates whether the concentration of market power actually causes negative abuse, and why disclosure will remedy this. The issue of causality is presupposed due to Whole Earth Network's experience with UUNet. (The role that can be played by disclosure of peering terms, for which he calls, is addresses later in this section.)

Moreover, it is unclear whether the 1996 Act applies to the Internet. Considered an "enhanced service" under the FCC, the Internet has traditionally been exempted from regulations. As Ferguson points out, this hands-off approach is maintained in the 1996 Act.34 Although Holub and others might argue that the intent of the Act cross-applies to Internet provision, the Act itself is not meant to place regulations on the Internet provider sector.

Nevertheless, the harms Holub identified from his experience creates a situation where a network may be held hostage because of the lack of market information and lack of alternatives for peering.

Alternatives and Information

A cornerstone of competitive markets is that alternatives exist for buyers and sellers. This is not entirely the case for Tier-1 ISPs. In cases where smaller-sized ISPs without national infrastructure require connectivity/transit from backbone ISPs, there exists a raft of options among the larger players to fill the market need. Their numbers may have shrunk over the past year due to acquisitions, yet there exists a number of choices, and as far as it can be observed, these national ISPs compete vigorously among each other.

But the issue is more complex for Tier-1 ISPs that control large percentages of the Internet. Here, the networks must exchange traffic or else a large proportion of users will not have connectivity to others, and positive network externalities—the value of the network in terms of its reach—would be severely reduced for all parties. Additionally, the large Tier-2 networks with national infrastructure are particularly pressed to become a customer of the Tier-1 network(s). The Tier-2 network's business model is to provide national carriage to customers and downstream ISPs, and peering is integral to their Internet connectivity. Because peering means they will be able to exchange traffic without paying a settlement, it is also financially vital. The removal of peering automatically places a Tier-2 network in the position of a customer in order to maintain offering Internet service.

Another principle of market economics comes into play in this regard: The reliance on information to base decisions. One of the key characteristics of peering arrangements by Tier-1 ISPs today—which was not the case a year ago—is that the criteria for peering is not disclosed. This effects the scenario between ISP-A and ISP-B cited above. Without knowing how other interconnection agreements have been struck, a massive game of "chicken" is currently underway, with many ISPs throwing around their weight and hoping the other ISP swerves first. To swerve, in this case, is to pay. And although the real cost may today be small, ISPs are more concerned with the symbolism the settlement would bring. Once delegated as a customer of backbone transit, ISPs well understand they will probably remain there. Also, creating settlement-based interconnections with one backbone very likely means the ISP will become the customer of other large networks seeking a settlement. Instead of peering with many providers, the ISP now is driven to become a full-fledged downstream reseller of one or more backbone providers.

For an ISP building out a national network, such as those in the Tier-2 category, the affects of this are disastrous. They are faced with an unpredictable commercial environment for investment and operations. Without a degree of market certainty for peering, which some sort of structure and public disclosure would provide, an important potential operational cost cannot be determined. This might discourage the creation of new market entrants.

While small players are benefited in a customer relationship for interconnections since they are provided transit and routability on the global Internet, the issue is more complex for mid-sized ISPs. The removal of a Tier-2 structure would place ever-more concentrated power in the hands of those ISPs that peer together. Not only do Tier-1 ISPs provide national carriage, they also dole out their customers' routes to other networks. A Tier-2 ISP might not need the infrastructure of the larger network to send traffic, but they do need to interconnect. The additional cost placed by the larger network on the smaller in this instance is not due to recovering the investment on sunk cost, but on the value of reaching customers on their network as a function of their market dominance. This, if it occurred so neatly, would constitute abuse of market position.

Thus the fundamental dilemma, otherwise stated, is that backbone providers control two resources: Not just the transit of traffic, but the accessibility of other networks to the end customers on their network.

Since a specific settlement model has not been established for backbone ISP interconnection and peering among other Tier-1 and Tier-2 ISPs, most players are caught playing arbitrage; seeking revenue from ISPs that they pass less as well as more traffic on to. By threatening to de-peer, ISPs risk "mutually assured destruction." The balance of power may be roughly equal among the Tier-1 ISPs, and this for now apparently serves to maintain stability.

Network Dynamics

By many accounts, these problems are not unique to the ISP sector, and have bedeviled networks of all types. Neuman, et al., for example, note the similarities among canal, rail, telegraphs and telephone networks. A commonalty so acute, they observe, that the 1934 Communications Act which regulated the U.S. phone system closely resembles the language of the 1887 Interstate Commerce Act which began federal regulation of the railway sector.35

With a historical approach to network evolution, the question arises whether the Internet needs to be similarly regulated as the telecommunications sector is currently. MacKie-Mason and Varian specify that the Internet is not regulated unlike common carriers, since the latter are natural monopolies.36 Yet regulation is not prima facie rejected. MacKie-Mason and Varian elsewhere generally believe a coordination role is necessary to preserve the Internet's network externalities. Speaking not of interconnection, but on the need for standards among networks to be imposed by "public and quasi-public bodies" for congestion control, accounting and usage-based pricing activities, they write: "One role for government is to insure interconnectivity between competing network providers."37

Indeed, if the notion of common carriage is applied to private Internet backbones (based on public interest, as Holub suggests), then government regulation of ISP interconnection would be justified.38 The principle of common carriage demands that networks provide service to all operators on an equal basis, may not refuse to sell to competitors, and must publish prices.

Brock, regarding telecom issues, believes there is a compelling social interest in regulating interconnections on grounds of market efficiency and competition.39 However, the network dynamics that comprise Internet service provision and the voice telecommunications sector are similar. As Brock explains: "Interconnection of ISPs and telephones is precisely the same thing in economic terms," there is "no substantive difference."

On the subject of network evolution and domination, Brock believes the Internet currently allows competitive forces to work, yet adds that the system is fragile if a single market leader emerges: "A network system tends toward monopolization whenever one company has enough of the market to conduct an effective business without interconnection with the other companies. The current Internet backbone structure has enough companies that no one of them can effectively operate without interconnection with the others. Any company in the current Internet that refused to interconnect with others would merely put itself out of business. On the other hand, if one company had a large enough share of the market that no other company could provide a useful s