2+2 tier banded frameworks of interconnectedness: industry structure determinants.

Shin, SeungJae ; Tucci, Jack ; Weiss, Martin B.H. 等

ABSTRACT

The Internet industry is generally considered to be vertically structured with the Internet Backbone Provider (IBP- long distance service carrier) in the upstream and Internet Service Providers (ISP) in the downstream. Although there are many ISPs and IBPs in each stream, both markets are considered independent oligopolies in that there are a few dominant providers for both ISPs and IBPs. The market leaders in each market create their own hierarchical tier and it is generally accepted that the Internet industry structure has evolved into a four-tier hierarchical structure. To understand the Internet industry, it is necessary to understand interconnection between ISPs and IBPs. The key element as an industry structural determinant is peering interconnection and the relationship created by that interconnectedness. Peering interconnection occurs within the same tier and the transit interconnection between the different tiers. This paper examines the internet industry structure using market share and interconnection strategies.

INTRODUCTION

The Internet industry is dynamic. It consists of millions of computers and switching devices. The number of Internet Service Providers (ISPs) increased rapidly from the mid 90' and the structure of the industry continues to change. It is widely accepted that today's Internet industry has vertical structure: over 40 Internet Backbone Providers (IBPs) including 5 top-tier backbones constitute the upstream industry (Kende, 2000) and over 10,000 ISPs for accessing the Internet make up the downstream industry (Weinberg, 2000). A backbone provider service is critical for any ISPs desiring to connect to the Internet. There are manifold interconnections between ISPs and IBPs. Moving data from one interconnection (tier-to-tier (IBP-to-ISP), or IBP-to-IBP, ISP-to-ISP) to another is the catalyst changing the Internet market structure. Compounding this is when users change mode of access from narrowband (dial-up) to broadband (DSL or Cable Modem). Narrowband dial-up access has been a major way to connect the internet, but in the summer of 2004, the number of broadband internet users surpassed the number of narrowband dial-up users. In this paper, we analyze the dynamic internet industry (both IBP and ISP) using market share and Internet interconnection strategies and dissect its complicated industry structure.

VERTICAL INDUSTRY STRUCTURE

Overview of Internet Industry

The Internet industry integrates the equipment, software, and organizational infrastructure required for Internet communications. As a rough approximation it can be said it is divided into two components: IBPs that transfer communications in bulk among network exchange points, and ISPs that (1) receive communications from individuals or institutions and transfer them to an IBP's network, and (2) receive communications from a IBP and transfer them to their destination. ISPs are used to refer to any company who can offer Internet connectivity. Some people use ISPs as a general term including IBPs. Some people argue that ISPs can be differentiated from other types of online information services, such as CompuServe or American On Line, because ISPs do not provide content but they focus only on providing Internet connectivity.

Generally speaking, the Internet industry has a vertical structure: Upstream IBPs provide an intermediate good and downstream ISPs using this input sell connectivity to their customers. Simplified, we suggest the analogy that the relationship between IBPs and ISPs is very much like that of "wholesalers and retailers."

In reality the internet is much more complex. The ISPs themselves are networks of users that may directly exchange information among each other. In addition, the IBPs may provide services directly to users and also may interconnect with other IBPs. A gestalt perspective gives the understanding that the internet is a network of networks that is accessible in many parts of the world. Since the telephone industry is tightly intertwined with the Internet industry, we begin by with its examination.

Telephone Industry and its Relationship with Internet Industry

Public Switched Telephone Network (PSTN) was designed and optimized for the transmission of the human voice. In the United States, telephone service is divided into two industries: (1) local telephone service provided by Local Exchange Carriers (LECs) and (2) long distance telephone service provided by Interexchange Carriers (IXCs). This structure creates a vertical hierarchy: Upstream IXCs provide the connection between LECs, and the downstream LECs have direct access to telephone users. The hierarchical structure of telephone industry has a strong impact on the Internet industry. This is due to the fact that many of these companies provide some type of service to either or both the ISP and IBP and/or the end user. End users in this case can be either public, private, or governmental users.

Traditionally, a LEC was a monopoly that served a specific geographic region without competition. Even after deregulation, LECs are still considered by many to be a local monopoly, especially for residential customers. In the U.S., the local telephone services provided flat-rate billing, that is, a telephone user can originate local calls as many times and as long as he wishes with only monthly flat rate. This type of billing system has been a great influence on the growth of Internet access market. This is not true for foreign countries where consumers often pay for time and distance for each call, whether it is local or long distance. Although not the topic of this paper, it is a good example why the internet is so expensive in countries where time and distance charges are levied against each call as compared to local telephone rates in the U.S.

The long distance market is now generally considered to be a very competitive market, though it too was a monopoly less than thirty years ago. Users can make a long distance call with pre-selected IXC through their LEC. Any IXC that wishes to handle calls originating in a local service area can build a switching office, called a Point of Presence (POP). The function of the POP is to interconnect networks so that any site where networks interconnect may be referred to as a POP.

Dial-up access using PSTN is the most universal form of Internet access. In the U.S., a modem call is typically a local call without a per-minute charge. ISP's lines are treated as a business telephone user not as a carrier, so they are not required to pay the measured Common Carrier Line Charge (CCLC) for originating and terminating calls, which recovers part of the cost of the local loop. The switching system in LEC's POP connects calls between Internet users and ISP's modem pool so the LECs' facilities support dial-up Internet communications. In addition, IBPs and large ISPs often construct their backbone networks by leasing lines from IXCs and LECs. As a result, we can say that telephone industry provides the basic infrastructure for the Internet industry.

INTERNET INTERCONNECTION STRATEGY

Background of Internet Interconnection

There are two types of Internet interconnection among ISPs and IBPs: peering and transit. The only difference among theses types is in the financial rights and obligation that they generate to their customers (Weiss & Shin, 2004). To understand the relationship between peering and transit, it is necessary to recall the non-commercial origin of the Internet. During the Internet's early development, there was only one backbone and only one customer, the military, so interconnection was not an issue. In the 1980s, as the Internet was opened to academic and research institutions, the National Science Foundation (NSF) funded the NSFNET as an Internet backbone. Around that time, the Federal Internet Exchange (FIX) served as a first point of interconnection between federal and academic networks. At the time that commercial networks began appearing, general commercial activity on the Internet was restricted by Acceptable Use Policy (AUP), which prevented the commercial networks from exchanging traffic with one another using the NSFNET as the backbone (Kende, 2000). In the early 1990s, a number of commercial backbone operators including PSINet, UUNET, and CerfNET established the Commercial Internet Exchange (CIX) for the purpose of interconnecting theses backbones and exchanging their end users' traffic. The NSF decided to stop operating the NSF backbone which was replaced by four Network Access Points (NAPs) (Minoli and Schmidt, 1998). NAPs are public interconnection points where major providers interconnect their network and these connections consist of high-speed switchs or a network of switches to which a number of routers can be connected for the purpose of traffic exchange. The function of NAPs is similar to major airport hubs; all ISPs and IBPs are gathered at the NAPs to connect each other. The NSF required that any ISP receiving government contracts or receiving money from public universities must connect to all of the NAPs. After the advent of CIX and NAPs, commercial backbones developed and a system of interconnection known as peering quickly evolved.

Peering Interconnection Strategy

The term "peering" is sometimes used generically to refer to Internet interconnection with no financial settlement, known as a "Sender Keeps All (SKA)" or "Bill and Keep," which can be thought of as payments or financial transfers between ISPs in return for interconnection and interoperability (Cawley, 1997). Peering can be divided into several categories: (1) according to its openness, it can be private peering or public peering, (2) according to the numbers of peering partners it can be Bilateral Peering Arrangement (BLPA) or Multilateral Peering Arrangement (MLPA), and (3) according to the market in which it occurs, it can be primary peering in the backbone market or secondary peering in the downstream market (Weiss and Shin, 2004). A peering arrangement is based on equality, that is, ISPs of equal size would peer. The measures of size could be (i) geographic coverage, (ii) network capacity, (iii) traffic volume, (iv) size of customer base, or (v) a position in the market. The ISPs would peer when they perceive equal benefit from peering based on their own subjective terms (Kende, 2000).

The original four NAPs were points for public peering. As the Internet traffic grew, the NAPs suffered from congestion. Therefore, direct circuit interconnection between two large IBPs was introduced (called bilateral private peering) which takes place at a mutually agreed place of interconnection. This private peering is opposed to public peering that takes place at the NAPs. It is estimated that 80 percent of Internet traffic is exchanged via private peering (Kende, 2000).

Before the Internet privatization, the NSF was responsible for the operation of the Internet. There are probably around 50 major NAPs world-wide in the internet, most of which are located in the U.S. (Moulton, 2001). As the Internet continues to grow, NAPs suffer congestion because of the enormous traffic loads. Because of the resulting poor performance, private direct interconnections between big IBPs were introduced, called peering points.

From the interconnection perspective, NAPs are the place for public peering. Anyone who is a member of NAP can exchange traffic based on the equal cost sharing. Members pay for their own router to connect to the NAP plus the connectivity fee charged by the NAP. Historically, in public peering, there was no discrimination for interconnection among the service providers (no priority given or taken based on usage). On the other hand, the direct interconnection between two equal sized IBPs is bilateral private peering, which takes place at the mutually agreed place of interconnection.

According to Block (Cukier, 1999), there are two conditions necessary for the SKA peering, that is, peering with no settlement, to be viable: (1) The traffic flows should be roughly balanced between interconnecting networks; and (2) the cost of terminating traffic should be low in relation to the cost of measuring and billing for traffic. The conclusion drawn from the above observations is that peering is sustainable under the assumption of mutual benefits and avoidance of costly, unnecessary traffic measuring. Nevertheless, peering partners would make a peering arrangement if they each perceive that they have more benefits than costs from the peering arrangement. Most ISPs in the U.S. historically have not metered traffic flows and accordingly have not erected a pricing mechanism based on usage. Unlimited access with a flat rate is a general form of pricing structure in the Internet industry. Finally, peering makes billing simple: no metering and no financial settlement.

Peering benefits come mainly from the network externality. Network externalities arise when the value or utility that a customer derives from a product or service increases as a function of other customers of the same or compatible products or services; that is, the more users there are, the more valuable the network is (Kende, 2000). There are two kinds of network externalities in the internet. One is direct network externality: the more E-mail users, the more valuable the internet. The other is indirect network externality: the more internet users there are, more web content will be developed which makes the internet even more valuable for its users. The ability to provide direct and indirect network externalities to customers provides an almost overpowering incentive for ISPs to cooperate with one another by interconnecting their networks (Kende, 2000). Contributing to the motivation for peering is lower latency because cooperation makes it necessary for only one hop to exchange traffic between peering partners.

Transit Interconnection Strategy

Transit is an alternative arrangement between ISPs, in which one pays another to deliver traffic between its customers and the customers of other provider. The relationship of transit arrangement is hierarchical: a provider-customer relationship. Unlike a peering relationship, a transit provider will route traffic from the transit customer to its peering partners. With transit agreements, usually small IBPs are able to receive and send communications using the facilities of large IBPs, and must pay a fee for these services. A concern related to transit is that while small IBPs do not have to pay in the case of peering through NAPs, they must pay a transit fee if they directly connect to one of the large IBPs. Before the commercialization of the Internet, carriers interconnected without a settlement fee, regardless of their size. However, after the Internet's commercialization, the large IBPs announced their requirements for a peering arrangement, and any carrier who could not meet those terms would be required to pay a transit fee in addition to the interconnection costs. An IBP with many transit customers has a better position when negotiating a peering arrangement with other IBPs. Another difference between peering and transit is existence of a Service Level Agreement (SLA), which describes outage and service objectives, and the financial repercussion for failure to perform. In a peering arrangement, there is no SLA to guarantee rapid resolution of problems. In case of an outage, both peering partners may try to resolve the problem, but it is not mandatory. This is one of the reasons peering agreements with a company short on competent technical staff are broken. In a transit arrangement it is a contract and customers could ask the transit provider to meet the SLA. Many e-commerce companies prefer transit to peering for this reason. A one minute outage cause the IBP, ISP, and the customer losses, hence, rapid recovery is critical to their business. Furthermore, in the case of transit, there is no threat to quit the relationship while in the case of peering a non-renewal of the peering agreement is a threat. ISPs are not permitted to form transit relationship over public NAPs because these are designed as a neutral meeting place for peering. When purchasing transit service, ISPs will consider other factors beside low cost: performance of the transit provider's backbone, location of access nodes, number of directly connected customers, and a market position.

ANALYSIS OF BACKBONE AND ACCESS MARKETS

Internet Backbone Market

With some simplification, it can be said that the IBPs receive communications in bulk from POPs or NAPs and distribute them to other POPs or NAPs close to the destination. To make the Internet a seamless network, the IBPs have multiple POPs distributed over the whole world. Most frequently they are located in large urban centers. These POPs are connected to each other with owned or leased optical carrier lines. Typically, these lines are 622 Mbps (OC-12) or 2.488 Gbps (OC-48) circuits or more, as defined by the SONET hierarchy, a standard for connecting fiber-optic transmission systems. These POPs and optical carrier lines make up the IBP backbone network. The IBP's POPs, are also connected with the POPs of many ISPs. The relationship between an ISP's POP and IBP's is the same as that of ISPs and IBPs.

According to Erickson (2001), the North American backbone market had around 36 IBPs in the first quarter of 2001. However, these numbers misinterpret the Internet backbone market structure because this market is highly concentrated. There were 11,888 transit interconnections between backbone and access markets in April 2000 (McCarthy, 2000). Counting by the number of connections to downstream market, MCI/Worldcom is a dominant player in the backbone market with 3,145 connections and Sprint is the second largest backbone provider with 1,690 connections and AT&T (934 connections) and C&W (851 connections) are the third and the fourth.

Several of the large IBPs are subsidiaries of large telephone companies such as AT&T, MCI/WorldCom, Sprint, etc. Since these companies own the infrastructure needed for telephone services, they are very favorably positioned to provide the facilities and equipment required by the IBPs. In addition, due to their economies scale, they are able to offer large volume discount rates or bundling agreements of both telephone and Internet lines for the services they provide. This is possible because the Internet industry is lightly, if at all, regulated. In particular, there are no regulations with respect to the tariffs that can be charged for the services provided. From these observations it follows that the large IBPs, supported by the large telephone companies, are in a position to capture large shares of the upstream market.

According to the Carlton and Perloff (1999), the most common measure of concentration in an industry is the share of sales by the four largest firms, called the 'C4' ratio. Generally speaking, if the C4 ratio is over 60, the market is considered a tight oligopoly. For the upstream backbone market this ratio based on the 1999 U.S. backbone revenue (Worldcom 38%, Genuity 15%, AT&T 11%, Sprint 9%) is 73, which shows high concentration in the market. The entry barrier is also high because there is a large sunk cost for nationwide backbone lines and switching equipment. The number of IBPs for the past three years shows just how high the entry barrier in the backbone market is: 43 (1999), 41(2000), and 36 (2001) (Erickson, 2001). The slight reduction for last three years is caused by mergers and acquisitions and reclassification. According to the number of players, we conclude that the overall backbone market is stable although oligarchic. In addition there are significant economies of scale and the rapid pace of technological change generates a large amount of uncertainty about the future return on investments. It is not easy to enter this market without large investments and advanced technology.

The interconnection price is usually determined by the provider's relative strength and level of investment in a particular area (Halabi, 2000). It is certain T-1 transit price has been decreasing continuously. In 1996, the internet connectivity for T-1 was $3,000 per month with $1000 setup fee (Halabi, 1997). According to Martin (2001), the average price of T-1 connection in 1999 was $1,729. In 2000, it was $1,348. In 2001, it was $1,228. One of reasons for decreasing T-1 interconnection price is advent of substitute services for T-1 line, such as wireless Internet access technology (LMDS, Satellite), digital subscriber line (DSL) technology, and cable-modem technology, which exerts a downward pressure on T-1 prices. As technology continues to improve and data transmission rates increase, pressure will continue to maintain a cap on service.

Internet Access Market

An ISP's product is public access to the Internet, which includes login authorization, e-mail services, some storage space, and possibly personal web pages. The ISP's coverage area is usually determined by the existence of an ISP's POP within the local telephone area. ISPs are classified as local, regional, and national according to the scope of their service coverage. The distribution of ISPs is presented in the Table 2.

Among 307 telephone area codes in U.S., the largest ISP covers 282 area codes and the smallest covers only 1 area code. The ISPs with 1 to 10 area codes constitute 79.81% of the total number of ISPs. This explains that most of ISPs in the downstream market are small, local companies. Some of these small ISPs are subsidiaries or affiliates of CLECs (Competitive Local Exchange Carriers), which are small telephone companies established in the 1990s as a result of telephone industry deregulation.

AOL-Time Warner is a dominant player in the dial-up access market. According to Goldman (2004), AOL-Time Warner had 22.8 million subscribers in the 3rd Quarter of 2004. AOL-Time Warner's subscribers are 22.8 million (24%) out of 81.1 million U.S. subscribers (Goldman, 2004). The Table 2 shows top 10 dial-up ISPs ranked by the number of subscribers.

In the downstream access market the C4 ratio is 43 and would be defined as relatively concentrated. However, the entry barrier in the downstream market is much lower than in the backbone market. Since subscribers can utilize the PSTN line to connect ISPs' modems and ISPs purchase business telephone lines from a LEC, ISPs for dial-up service do not have to invest in access lines to individual subscribers. They can build POPs to link to the PSTN and other ISPs. Since a T-1 lines prices and telecom equipment prices are currently dropping quickly, a large number of small ISPs are possible, especially in the less densely populated areas. The number of North American ISPs for the past several years is an evidence of low entry barrier in the downstream market: 1447 (February 1996), 3640 (February 1997), 4470 (February 1998), 5078 (March 1999), 7463 (April 2000), and 7288 (March 2001) (Erickson, 2001).

Most ISPs provide unlimited Internet access with a monthly flat rate. For major national ISPs, the price ranges generally from $0 to $25 per month dependent on the level of service. Some ISPs provide Internet access service with zero monthly subscription fees to their customers; their revenues depend completely on Internet advertising income. Some base their service on speed, while others on memory usage as an upgrade to their standard service.

ISPs are free to make local peering arrangements with other ISPs. Cremer and Tirole (2000, p445) call this local secondary peering. The Pittsburgh Internet Exchange (PITX) is an example of local peering arrangement. Without this local peering, all network traffic passing from one Pittsburgh network to another had to be sent through Washington, D.C., Chicago, or New York City. The sending and receiving networks pay an unnecessary cost for this inefficient handling of data that should have remained local. Participants in this local exchange point reduce their costs and improve performance and reliability for their local Internet traffic with the equal basis of cost recovery. However, this kind of peering is confined to local internet traffic. Outbound traffic (connected to other networks through an IBP) to other areas still has to depend on the IBP's transit service.

Broadband Internet Market

The Internet access technologies are roughly divided into two categories: narrow band access using dial-up modem technology and broadband access such as Cable-Modem, DSL, and wireless broadband access technology. Among the above broadband access technologies, DSL and Cable-Modem are the two dominant broadband access methods. According to Vara (2004), in July of 2004, more than half of U.S. internet users connected to the internet using a broadband service. It was the first time high-speed broadband internet connection had more market share than dial-up connection. The broadband service providers usually confine their business to high density regions because broadband service requires large investments in "advanced" (read expensive) technologies. Internet users in the rural area rarely have a chance to enjoy the benefit of the higher speed access that broadband services offer.

According to Leichtman Research Group, the twenty largest cable and DSL providers in the US account for about 95% of the market in high speed internet access. The top broadband providers now account for over 30.9 million high-speed Internet subscribers, with cable having nearly 18.8 million broadband subscribers, and DSL trailing behind at 12.2 million subscribers. If we confine the access market into the broadband technology, the C4 ratio in this market is 55% which is considered oligarchic. The following table shows the top 10 broadband access providers in the U.S.

DE-PEERING AND FOUR-TIER HIERARCHICAL STRUCTURE

In 1996, AGIS was the first IBP to unilaterally terminate peering arrangements. After that, a series of IBPs announced that they were ending peering with many of their previous peering partners and were no longer accepting peering arrangements from other networks whose infrastructure would not allow the exchange of similar levels of traffic access. Instead of peering, they would charge those smaller ISPs for transit. Finally, the large IBPs moved away from public NAPs to private peering or maintained relatively small capacities like T3 in the NAPs and then placed themselves in a new hierarchy, so called top-tier IBPs (Jew and Nicolls, 1999). Most top-tier IBPs are subsidiaries or affiliates of the major facilities-based telecommunication carriers. They are UUNET (Worldcom), C&W, Genuity, AT&T WorldNet, and Sprint, the 'so called' Big 5. They don't need transit service from others and they make peering arrangement with each other. Over 80% of the U.S. backbone traffic is estimated to pass through their systems and switches (Weinberg, 2000). Other non Big 5 IBPs make peering arrangements among themselves and simultaneously purchase transit services from the Big 5.

There are two cases for which peering is generally refused: (1) Regional IBPs which do not have a national backbone network and (2) content providers or web hosting companies, so called web farms. The main reason for this refusal is a free-rider issue. Peering partners generally meet in a number of geographically dispersed locations. In order to decide where to pass traffic to another, they have adopted what is known as "hot-potato routing," where an ISP will pass traffic to another backbone at the earliest point of exchange. Under the hot-potato routing rule, someone who does not have a national backbone network must transport its traffic on the others' backbone networks. In addition to that, asymmetric traffic patterns, which occur in file transfer or web surfing, result in increased capacity costs without commensurate revenues.

Some of the Big 5 recently disclosed their policy for peering but some of them still do not. There is an unwritten rule shared by the Big 5 about their peering standard: (i) A coast to coast national backbone with a certain level of bandwidth requirement, (ii) a number of presences in the major exchange points, (iii) 7 days by 24 hours Network Operation Center (NOC) and highly experienced technical staffs, and (iv) a certain level of traffic ratio between inbound and outbound, usually 1:4. It is indeterminate what the exact requirements for the private peering are since peering agreements are under non-disclosure. Without a doubt, these requirements could be a significant entry barrier for any new entrant.

PSINet, which was one of the large IBPs, used a peering standard called "open peering policy", that was different from the Big 5. It would peer with any ISP including local, regional, and national except for companies whose primary business was web hosting or content collection. Some of the Big 5 did not want to peer with PSINet, because some of PSINet's private peering partners are transit customers of the Big 5. Whenever the Big 5 upgrade their networks, they upgrade their peering policy. From the tier-2 IBP's point of view, peering requirements are getting tougher and tougher. Nobody can enter into the top tier group without their approval. This cartel-like behavior has been an important issue in the Internet industry for several years and will eventually be a sticking point with the Federal Trade Commission in the future.

After being refused peering in the backbone market, ISPs in the downstream access market, usually operating in a limited geographic region, tried to peer among themselves. Cremer and Tirole (2000) in their paper call this kind of peering "local secondary peering." This is a major factor in proliferation of local and regional Internet exchange points. These smaller exchange points (compared to NAPs) referred to as Metropolitan Exchange Points (MXPs) (OECD, 1998).

Most of Internet exchange points, or POPs of major IBPs are located near the metropolitan areas, which are far from the rural areas. The local ISPs in a rural area have to lease long expensive lines to reach an interconnection point. The long distance from private or public peering points is an additional obstacle to overcome for the rural ISPs. Through de-peering in one tier and peering in lower tier in both markets, the four-tier hierarchical structure has emerged; in the backbone market, tier-1 IBPs with their nationwide backbones interconnect each other and make a core network in the Internet and tier-2 IBPs with their regional backbones interconnect each other and pay tier-1 IBPs for connectivity to rest of internet, which mean they are customers of tier-1 providers. A few of nationwide big ISPs are also a member of tier-2 group. In the access market, tier-3 regional ISPs are customers of tier-1 or tier-2 connecting them to the rest of internet. Local, small ISPs are tier-4 providers and they are customers of higher tier providers. However, the demarcation between the tiers is not clear. In the following section we explain how peering decisions are made.

PEERING DECISION-MAKING PROCESS

An interconnection strategy may be different according to its priority. If expense of interconnection is the number one issue, ISPs will try to find as many peering partners as they can and try to choose minimum combination costs among them. Or if performance is the top priority, they may prefer private peering or transit to public peering. All interconnection decisions start from the analysis of their own traffic. An ISP should try to find the available options and negotiate with their interconnection partners for interconnection methodology, interconnection line capacity, interconnection settlement, etc. This process will be explained below in detail. (Norton, 1999)

Phase I: Identification of ISP's Traffic

The costs of peering and transit vary according to the distance of the ISPs' POP and interconnection point. Generally, the cost of transit is more expensive than that of peering. Before deciding on a transit or peering arrangement, the ISP may systematically sample inbound and outbound traffic flows and then map these flows to the originating Autonomous System (AS), which is defined as a collection of networks that are under the administrative control of a single organization and that share a common routing strategy. Calculations are made to determine where to reduce the load on the expensive transit paths.

Phase II: Finding Potential Interconnection Partners

Based on the traffic map and the aforementioned analysis, ISPs try to find interconnection partners. Because peering policies are often exposed only under Non-Disclosure Agreements (NDA), it is not easy to know them in advance of negotiations. It is reasonable for an ISP to find its peering partners in its own level of Internet industry hierarchy. If a top-tier IBPs makes a peering arrangement with a second tier IBP, then the latter could be the formers' competitor. Therefore, a higher tier ISP would prefer selling transit service to lower tier ISPs and have an incentive to reduce the number of their own competitors. Many ISPs, except for top tier IBPs, have adopted a hybrid approach to interconnection, peering with a number of ISPs and paying for transit from one or more IBPs in order to have access to the transit provider as well as the peering partners of the transit provider.

Phase III: Implementing Interconnection Methodology

Since peering is seen as being of mutual benefit, both parties explore the interconnection methods that will most effectively exchange traffic. Both parties decide (1) how many interconnection points they have, (2) where to locate the interconnection points, (3) how they interconnect, private peering or public peering, (4) what line capacity they will use, (5) settlement free or settlement involved, etc.

Table 6 illustrates comparison of per Mega-bit cost of transit, private peering, and public peering. If we compare cost per Mbps shipped (CPMS) per month of OC3 capacity, the order from the cheapest is public peering ($30), private peering ($64~$129), and transit ($464).

CONCLUSION

The Internet has become an important social and business tool. Furthermore, the market has become even more dynamic since it was privatized. Peering has emerged as a phenomenon that can at one time be beneficial to both parties while simultaneously discriminate against one of the peering partners. Professor Frieden calls it a "balkanization" in the Internet. If a new technology was introduced in this market, the internet providers with this technology would have a tendency from past practices to create their own peer groups to make money against the provider without this technology. On the other hand, ISPs are competitors and cooperators simultaneously: competitors for market share and cooperators for global connectivity. One ISP's decision has an influence on other ISP's decisions. Thus, they have a strong dependence on each other beyond just competitive factors.

In our paper, we believe that the two-tier, two layer market structure of both backbone and access is oligopolistic. This means, if a new technology is developed lowering costs, or increasing speed, or if some of them reach an agreement they could exercise their market power, maybe for the better for the consumer, maybe not. A policy maker's goal for the Internet industry is continuing growth and innovation. To achieve this goal, regulators need to continue to encourage competition and to give incentives for ongoing investment and in the development and deploying of new technologies, which will benefit consumers in the internet market. Therefore, it is a role of internet policy makers to make socially desirable competitive environments between higher tier ISBs and lower tier ISPs in the Internet industry.

REFERENCES

Carlton, D. & J. Perloff, (1999). Modern Industrial Organization, 3rd Edition. New York:Addison Wesley Longman.

Cawley, R.A. (1997). Interconnection, pricing, and settlements: Some healthy jostling in the growth of the Internet. In Kahin, B. & J. Keller (1997). Coordinating the Internet (pp. 346-376). Boston, Massachusetts: MIT Press.

Cremer, J., P. Rey, & J. Tiroel, (2000). Connectivity in the Commercial Internet. The Journal of Industrial Economics, 40(8), 433-472.

Erickson, T., (2001). Introduction to the Directory of Internet Service Providers, 13th Edition. Boardwatch Magazine, from http://www.ispworld.com/isp.

Frieden, R. (1998). Without public peer: The potential regulatory and universal service consequences of Internet Balkanization. 3 Virginia Journal of Law & Technology 8, from http: //vjolt.student.virginia.edu/.

Goldman, A., (2004). Top 23 U.S. ISPs by Subscriber: Q3 2004. ISP-Planet magazine, from http://www.isp-planet.com/research/ranking/usa.html

Greene, B.R., (2000). L2 Internet exchange Point (IXP) using a BGP Route Reflector. Draft version 0.4, from http://www.cisco.com/public/cons/isp/ixp.

Halabi, B., (1997). Internet Routing Architectures. Indianapolis, IN: Cisco Press.

Halabi, B., (2001). Internet Routing Architectures 2nd Edition. Indianapolis, IN: Cisco Press.

Jew, B. & Nicholls, R., (1999), Internet Connectivity: Open Competition In the Face of Commercial Expansion. 1999 Pacific Telecommunications Conference.

Kende, M., (2000). The Digital Handshake: Connecting Internet Backbones. FCC-OPP Working Paper No. 32, from http://www.fcc.gov/Bureaus/OPP/working_papers/oppwp32.pdf.

Martin, L., (2001). Backbone Web Hosting Measurements. Boardwatch Magazine, from http://www.ispworld.com/isp/Performance_Test.htm.

Minoli, D. & A. Schmidt, (1999). Internet Architectures. New York, NY: Wiley.

Moulton, P., (2001). The Telecommunications Survival Guide. Upper Saddle River, NJ: Prentice-Hall.

Norton, W.B., (1999). Internet Service Providers and Peering, draft version 1.9, from http://www.nanog.org/mtg-0010/tree.doc.

OECD, (1998). Internet Traffic Exchange: Development and Policy, from http://www.oecd.org/dataoecd/11/26/2091100.pdf

Vara, V., (2004). High-Speed Surpasses Dial-up As Top Home Web Access in U.S., from http://www.wsj.com.

Weinberg, N., (2000). Backbone Bullies. Forbes, June 12 edition, 236-237.

Weiss, M., & S. Shin, (2004). Internet interconnection economic model and its analysis: Peering and settlement, Netnomics, 6, 43-57.

SeungJae Shin, Mississippi State University--Meridian

Jack Tucci, Mississippi State University--Meridian

Martin B. H. Weiss, University of Pittsburgh

Hector Correa, University of Pittsburgh *

ENDNOTES

* Dr. Correa, an international scholar and professor in the Graduate School of Public and International Affairs at the University of Pittsburgh who passed away in the summer of 2004. We deeply appreciate his effort in this paper.

Table 1: Distribution of ISPs by their coverage
(Erickson, 2001)

Telephone area codes
covered by ISP Percentage Type

1 35.14% Local
2-10 44.67% Local / Regional
11-24 4.11% Regional
25-282 16.08% National

Table 2: Top 10 U.S. ISPs (Q3 2004)
(Goldman, 2004)

Rank & ISP Subscribers Market Share

(1) AOL 22.8M 24.0%
(2) United Online 6.6M 6.9%
(3) Comcast 6.5M 6.8%
(4) EarthLink 5.2M 5.7%
(5) SBC 4.7M 4.9%
(6) Road Runner 3.9M 4.1%
(7) Verizon 3.3M 3.5%
(8) Coax 2.4M 2.5%
(9) BellSouth 1.9M 2.0%
(10) Charter 1.8M 1.9%

Table 3: Top 10 U.S. Broadband ISPs (Q3 2004)
(Lehichman Research Group, http://www.itfacts.biz)

Rank & ISP Subscriber Market Share

(1) Comcast (Cable) 6.5M 20%
(2) SBC (DSL) 4.7M 14%
(3) Time Warner (Cable) 3.7M 11%
(4) Verizon (DSL) 3.3M 10%
(5) Cox (Cable) 2.4M 7%
(6) Bell South (DSL) 1.9M 6%
(7) Charter (Cable) 1.8M 6%
(8) Adelphia (Cable) 1.3M 4%
(9) Cablevision (Cable) 1.3M 4%
(10) Qwest (DSL) 1.0M 3%

Table 4: Per M-bit Cost Comparison
(AT&T (Transit), Norton (2000, Private Peering), and Chicago
NAP (Public Peering))

Interconnection
Type Capacity Cost / Capacity Per M bit Cost

Transit DS3 $26,000/45M $578/M
 OC3 $72,000/155M $464/M

Private Peering OC3 ($10,000~$20,000)/ $64/M ~ $129/M
 155M/2
 OC12 ($20,000~$30,000)/ $32/M ~$48/M
 622M/2
Public Peering DS3 $3,900/45M $87/M
 OC3 $4,700/155M $30/M