EPUBs are an experimental feature, and may not work in all readers.

Presented at MIT Workshop on Internet Economics March 1995

Abstract

We explore the economic factors associated with multiple Internet interconnection architectures, which have enabled the Internet to grow with the number of applications, number of users, and amount of network traffic. There are three Internet interconnection architectures identified by the paper: bilateral agreement, cooperative agreement, and third party administrator. While all are technically feasible, there are economic reasons for some companies to prefer one architecture over another. We argue this based on incomplete contract theory coupled with the ability of resale for interconnection. The paper concludes that bilateral agreements may be best for large Internet network providers while cooperative agreements may work best for smaller network providers. Third party administrators may exist, but the ability for the connected networks to resell their service, along with the ability for the administrator to act opportunistically, will deter this type of interconnection agreement.

Introduction

The growth of the Internet can be explained from the standpoint of technological development, public policy, or economics. While each of these elements could be identified as the main reason for Internet growth, it is likely that the three have played a complementary and reinforcing role. The developments in networking and computer technology have allowed faster and lower cost components to drive the system. Developments in software have led to new applications, making the infrastructure more useful. Meanwhile, the cost of these components have been dropping on a cost curve similar to that depicted by Moore who predicted that the ratio of computing performance to cost will double every eighteen months. Government support of the Internet through government agencies such as the National Science Foundation, NASA, and the Department of Energy have helped create a critical mass of users connected to the Internet. Finally, the policy of openness regarding standards and interconnection agreements has aided the process as well. These are just a few of the developments that have led to the growth of the Internet; there are too many to recount here.

These reinforcing developments provide the backdrop for the interconnection agreements that have created the Internet's "network of networks." Operation of the NSFnet as the "backbone" of the Internet made interconnection relatively clear: connect with a service provider who connects you to the NSFnet. Recently, with the movement from the NSFnet to many wide area networks making up the Internet backbone, people have been more concerned with connecting to the Internet's network of networks—the Internet cloud. A cloud is a perfect analogy for the wide area networking capabilities of the Internet since there is no fixed form and no undergirding structure that keeps it in place. Lack of a dominant Internet service provider has reinforced this cloud. Table 1 shows the fragmented Internet service provision market.

Table 1: 1994 Internet Service Provider Market Size [1]
Without Information Providers With Information Providers
Company Revenue (in millions) U.S. Market Share Company Revenue (in millions) U.S. Market Share
UUNET $46.78 9.52% CompuServe $216.0 20.76%
NETCOM $31.22 6.35% AOL $180.0 17.30%
Sprint $29.94 5.74% Prodigy $90.0 8.65%
PSI $22.92 4.67% UUNET $46.78 4.50%
Supernet $10.37 2.11% Delphi $44.80 4.31%
ANS $10.30 2.10% NETCOM $31.22 3.00%
CERFnet $8.0 1.63% Sprint $29.94 2.80%
BBN $6.82 1.39% PSI $22.92 2.20%
World $4.75 0.97% Supernet $10.37 0.94%
SURAnet $4.35 0.89% ANS $10.30 0.91%
Other $320.59 64.63% Other $361.47 34.63%
Total $496.04 100% Total $1,043.80 100%

With the fragmentation of the Internet's wide area network service market, interconnection between one Internet network and another becomes possible through many configurations. Each of the interconnection architectures have different technical, policy, and economic costs and benefits.

In this paper, we explore three interconnection agreement architectures from an economic, technical, and policy viewpoint. We define these models and assess the economic scalability of the systems to support continued Internet growth. This growth is evident in the amount of traffic, the number of users, and the number of applications. We conclude that this growth may be sustainable only with some types of interconnection agreements. This conclusion stems from analysis of the incentives for Internet interconnection agreements, interconnection architectures, and policies for pricing and resale.

Incentives for Interconnection

Perhaps the dominant driving economic forces for interconnection are positive network externalities. A network externality is the cost or benefit the user of a network derives from an additional person using the same network. The network externality is positive when the additional person represents a benefit to users while the network externality is negative if the additional user represents a cost. Theoretical economic analysis on network externalities has been done elsewhere (Katz and Shapiro, 1985; Farrell and Saloner, 1985) but there has been little work to find empirical evidence to support these claims. The Internet would be an excellent example where the positive network externalities drive the economics.

The network externalities are best seen when looking at the Internet at an application level. For example, electronic mail over an infrastructure is only as good as the number of people it can reach (having only one person on a network isn't very useful since you cannot send electronic mail to anyone). It may be possible that some applications lead to negative network externalities. Since Internet resources are shared among the users of the Internet, an additional user is a source of network congestion. If a network resource—whether an Internet server or bandwidth—were available to only one person, the user would incur a congestion cost resulting from another person connecting to the infrastructure. The positive network externalities, however, dominate the Internet and are a motivating factor for networks to interconnect. By interconnecting, they can reach more people and more people can reach them.

Through interconnection, users can achieve interoperability of their applications. Interoperability has been defined as "the ability of two or more systems (such as devices, databases, networks, or technologies) to interact in concert with one another, in accordance with a prescribed method, to achieve a predictable result; the ability of diverse systems made by different vendors to communicate with each other so that users do not have to make major adjustments to account for differences in products or services; and compatibility among systems at specified levels of interaction, including physical compatibility." [2]

Interoperability allows heterogeneity of technologies while allowing users to work together (as the definition suggests). The difficulty of interoperability with respect to interconnection agreements is that since interoperability is defined at the application layer and networks are interconnected at the IP layer, different interconnection architectures may affect interoperability differently. The technical and economic effect of this distinction is not explored in this paper. We define interoperability as the benefit the user derives in order to differentiate it from interconnection.

The implication of the importance of interoperability is that interconnection agreements may be application driven. Therefore, when designing network protocols, engineers must consider the interoperability implications at the application layer. Furthermore, by taking an application view of interoperability, economists, who traditionally have looked at applications and their effect on economic measures such as productivity, may be able to study the cost and benefits of the Internet. The difficulty with designing interconnection agreements is that applications on the Internet are dynamic—new applications are constantly being developed.

If the Internet matures into an infrastructure with dominant market players, interconnection may no longer be consistent with a business' competitive strategy. For example, if the number of users levels off, a business may try to lure customers away from another Internet service provider rather than trying to attract non-users. This company, Company A, may choose not to interconnect with another service provider, Company B, to entice customers to switch from company B to A in order to benefit from the content that resides on A's network. This is not very different than the telegraph and telephone networks developed in the United States in the nineteenth century. [3] Regulation may then be necessary for the Internet service provision market to require companies to interconnect. While we do not believe that the analogy between the telephone and the Internet is water-tight, it would be erroneous to think the U.S. government will never regulate interconnection agreements for Internet service provision.

Interconnection Architectures

The interconnection agreements we explore fall into three models, which we call architectures: bilateral agreements, cooperative agreements, and third party administrators. These architectures are both technical and economic in nature. The technical component is how many lines enter a node. The bilateral agreement has two lines entering a node so the routing can be done by one router with two inputs. The cooperative agreement and the third party administrator have more than two inputs. Therefore these two architectures usually deploy a high speed local area network between routers at the end of each terminating node. The economics of these architectures can be thought of as firm boundaries. The bilateral and cooperative agreements have ownership of the interconnection node shared between the firms. The third party administrator is a separate firm, which may be distinct from all of the other firms connecting to it.

The different architectures exhibit different levels of opportunism for the interconnected firms. For example, a firm that is able to exercise control over the interconnection point may design the architecture or collect data that would give them an advantage over the interconnected firms. The third party administrator architecture enables the firm controlling the interconnection node to exhibit opportunism, which is most evident. An example of this may be the Commercial Internet eXchange (CIX) which had difficulty collecting fees because of the ability for resale. Resale is an important issue when considering interconnection agreements and will be discussed later in this paper.

The coordination costs are the last economic factor discussed in this paper. Coordination between parties may be difficult especially when a firm has an opportunity for opportunistic behavior. Interconnection agreements that include a few firms (bilateral agreements) may not have high coordination costs while many interconnected firms (cooperative agreements) may have relatively higher coordination costs. The third party administrator architecture is similar to a market approach where firms have a choice in interconnection agreements but cannot prevent the third party administrator from acting opportunistically.

Bilateral Agreement

Perhaps the most basic interconnection architecture involves two networks that exchange data directly. Both firms benefit from interconnecting due to the network externalities. Since there are only two firms interconnecting, the coordination costs are low and there is little chance of opportunism because of the trust that has developed between the two firms. The architecture is relatively simple, as is the standards process.

Cooperative Agreement

The cooperative agreement involves many firms interconnecting at a node with a shared responsibility for the interconnection point. This architecture is similar to the Federal Internet eXchange (FIX) located on the East and West Coasts of the United States. The FIX interconnects government agencies such as the Department of Energy and NASA and coordinates their architecture through fora such as the Federal Networking Council. There is a shared vision among these firms, which makes the coordination costs lower since there is little opportunity for opportunism. The coordination costs may still be high, however, in setting standards for the interconnection node.

Third Party Administrator

The third party administrator approach to the interconnection problem is a market driven approach. This third party can act opportunistically and the interconnected firms may voice their concern in the third party's policies, but may only exit their contract as a mechanism to voice disapproval. The third party is often thought of as a common carrier since it does not discriminate between users. The coordination costs are less than the cooperative agreement with the same number of firms, it is assumed, since each firm must only interact with the administrator and not all other firms. There is a vulnerability in this architecture which may result when the interconnection point is dominated by one or a few firms. Dominance will affect the administrators' decisions since they do not want to lose the dominant firm as a user. Therefore, the firms that pay the most to the administrator may act opportunistically by influencing the way the administrator operates.

Resale and Usage-Sensitive Pricing

Perhaps the most critical component of opportunism with Internet interconnection agreements is the ability of a firm to resell its interconnection service. For example, firms A and B may want to interconnect at a node but they do not want to each pay the interconnection fee. They may decide to aggregate their traffic before it reaches the node. Firm A might also decide to resell the interconnection service to firm B at a price less than the interconnection agreement. Resale issues affect all three architectures discussed in the previous section. We will explore two mechanisms to prevent resale: pricing and policing.

Pricing issues provide incentives or disincentives for resale. For example, a per-bit charge may be assessed for the traffic exchanged at an interconnection node. This is a usage-sensitive pricing scheme, which is different than flat rate pricing usually embraced by Internet users. [4] Flat rate pricing schemes may also deter resale if the flat rate prices are consistent with the bit-rate of the interconnection link. For example, if a flat rate is charged to interconnecting firms at a dollar per megabit per second, firms A and B may still aggregate their data but they will experience latency costs unless they increase the bandwidth of the link which will, in turn, cost more.

If a policy not allowing resale is enforced at the interconnection node without using pricing as an incentive, policing is another option. This is similar to the model used by the Commercial Internet eXchange (CIX) where a flat-fee was charged to users as long as they didn't resell their service. This is enforceable by users specifying the allowable Internet Protocol (IP) addresses, which can send and receive information over a link. If an IP address outside of the allowable range tries to use this interconnection point, it will be blocked. Policing is very difficult and some users may be able to fool such a blocking mechanism by encapsulating an IP packet inside a legal IP address. Another cost of policing, which is also true with pricing, is the increased administrative costs.

The overhead associated with policing or pricing may increase the cost of interconnection. For example, Bailey, Gillett, et al (1995) report that the billing overhead for phone calls account for approximately fifty percent of a bill. The cost of the additional hardware and operations support for policing may be expensive as well. Both of these costs are realized when firms act opportunistically, which is the case with third party administrator architectures. The bilateral and cooperative agreement architectures involve a higher degree of trust and a consistency in goals, which may result in a policy that does not need the cost of enforcement.

Conclusion

A large Internet service provider is more concerned about its reliability and interconnection agreements since its users derive more interoperability benefit from interconnection. Therefore, it would be in the best interest of such providers to establish interconnection firms which they trust not to act opportunistically. This coordination cost would increase if they tried to enter into cooperative agreements. Furthermore, the third party administrator architecture would involve high vulnerability costs associated with the administrator acting incompetently or opportunistically.

Smaller networks with shared goals may best be served through cooperative agreements. Since the coordination costs are higher than the bilateral agreement architecture, a shared vision, such as the vision of the U.S. government agencies that operate Internet networks, reduces the coordination hassles. Furthermore, the shared vision will lead to agreements based on trust between the interconnected firms for a policy that addresses resale without having to implement usage-sensitive pricing or policing.

The role of third party administrators will most likely not diminish as the Internet grows. However, the ability for firms to resell this service coupled with the ability of the administrator to act opportunistically does not ensure that this may continue to be a good model. In favor of the third party administrator is the argument resulting from coordination costs, which are lower when firms can coordinate through a hierarchy.

Finally, we believe that many of the arguments made here could be supported by traffic analysis which extends beyond what is currently being done. While more data is being collected, the information is not consistent among different interconnection nodes. A repository of comparable data would help academics understand the pricing issues and formulate new contracts for interconnection architectures. Perhaps through interoperable and scalable interconnections the Internet will be able to meet the growing demand of users, applications, and traffic.



The author wishes to thank Lee McKnight, Erik Brynjolfsson, and David Clark for their contribution to this paper's content.


The author can be reached by email (bailey@rpcp.mit.edu) or at 617/253-6828, MIT Research Program on Communications Policy, 77 Massachusetts Avenue, MIT Building E40-236A, Cambridge, MA 02139. He is a doctoral candidate in the Technology, Management and Policy Program and a research assistant with the MIT Research Program on Communications Policy, Center for Technology, Policy, and Industrial Development.


References

Bailey, J., S. Gillett, D. Gingold, B. Leida, D. Melcher, J. Reagle, J. Roh, R. Rothstein and G. Seale (1995). "Internet Economics Workshop: Workshop Notes." MIT Research Program on Communications Policy, Massachusetts Institute of Technology.

Bailey, J. P. and L. W. McKnight (1995). "Internet Economics: What Happens When Constituencies Collide." INET '95, Honolulu, June.

Brock, G. W. (1981). The Telecommunications Industry: The Dynamics of Market Structure. Cambridge, MA, Harvard University Press.

Farrell, J. and G. Saloner (1985). "Standardization, Compatibility, and Innovation." Rand Journal of Economics, 16, Spring.

Katz, M. L. and C. Shapiro (1985). "Network Externalities, Competition, and Compatibility." American Economic Review, 75.

Kerrey, B. (1995). "Communications Interoperability Act of 1995." U.S. Senate, S.710.

Maloff, J. (1995). "1994-1995 Internet Access Providers Marketplace Analysis." The Maloff Company.


Notes

1. Maloff (1995), pp. 7-8.return to text

2. Kerrey (1995), p. S5452.return to text

3. See Brock (1981) for details.return to text

4. Bailey and McKnight (1995) discuss many reasons why the Internet has accepted flat rate pricing and not usage-sensitive pricing, save a few cases. The paper also details differences in terminology between flat rate, usage-based, and usage-sensitive. They argue that usage-based pricing is a term with a poor definition and will not be used in this paper. Usage-sensitive pricing exists when the marginal monetary cost for sending a bit is non-zero.return to text