Margit Vanberg, History of the Internet hierarchy in:

Margit Vanberg

Competition and Cooperation Among Internet Service Providers, page 42 - 45

A Network Economic Analysis

1. Edition 2009, ISBN print: 978-3-8329-4163-5, ISBN online: 978-3-8452-1290-6

Series: Freiburger Studien zur Netzökonomie, vol. 14

Bibliographic information
42 The public Internet With the termination of the NSFNET in April 1995 the connectivity between industrial users and academic networks was no longer hampered by the restrictions for using the NSFNET. Commercial networks took the place of the NSFNET in providing long-distance data transportation and connectivity services. Without the government restrictions on the type of traffic allowed, the Internet became public in the sense that all users could access and correspond with any site available on the public Internet by establishing the relevant interconnections. Universal connectivity In the commercial era of the Internet (1995 to the present), Internet users have come to expect that all sites and all users connected to the public Internet are accessible to them, irrespective of the home network providing the Internet access. This ability to reach all users and sites on the public Internet is called universal connectivity. Economides (2005: 389) notes: “The demand for universal connectivity on the Internet is stronger than the demand of a voice telecommunications customer to reach all customers everywhere in the world.” He argues that a long-distance company that does not offer access to all countries can survive in the market for long-distance services, because users can generally anticipate whether they will need to contact another user in a certain country. “On the Internet, however, one does not know where content is located. […] customers would never be able to know or anticipate what content they would be missing” (ibid.). Therefore, they will not be willing to accept a limited interconnectivity service. 3.3 History of the Internet hierarchy In the NSFNET era, the NSFNET was the single wide-area backbone network of the emerging “network of networks.” Public funding of the NSFNET backbone crowded out investments into higher network levels, because regional and local networks had no incentive to establish direct interconnections with more distant networks, since using the NSFNET backbone ensured connectivity with all other networks attached to the NSFNET at lowest possible costs. Later on, when ISPs began investing into direct interconnections between their networks in order to offer data transportation services to commercial users, the privatization design of the NSF continued to endorse the network hierarchy that had existed previously. Financial support could be received from the government for realizing interconnection with a commercial ISP, if this ISP had a backbone network that connected to all NSF-specified priority Network Access Points (NAPs) (NSF, 1993: 12).27 In consequence, only ISPs with a nation-wide network, reaching from the East Coast to the West Coast of the United States, and with points of interconnection in the central states were competitive in 27 At the time of the solicitation, the NSF had specified the locations California, Chicago, and New York City as priority locations. 43 offering interconnection services to regional networks. Such large investments were undertaken by only a few large carriers, preferably those that already had experience with large telecommunications networks. This led to the top-tier of the private Internet being made up of a few commercial ISPs with networks of wide geographic coverage. Figure 3.3 is a stylized depiction of the commercial Internet organization soon after the privatization of the NSFNET. Figure 3.3: The design for the privatization of the NSFNET Source: Kesan and Shah. 2001: 116 Besides these historical influences, there are also very practical reasons for the hierarchical organization of the Internet. If all networks making up the Internet were to establish direct interconnections with one another, then, given n networks, the number of interconnections required would be n(n-1)/2. With many thousand active ISPs, it is obvious that a direct interconnection between each and every operating ISP is not feasible.28 It is important to exploit economies of bundling, by, for in- 28 Marcus (2006c: 20) estimates the number of networks making up the Internet at around 30,000 to 40,000. Regional Network Regional Network Regional Network Regional Network Commercial Backbone Commercial Backbone Commercial Backbone NAP NAPNAP Local networks 44 stance, interconnecting to network exchange points where interconnections with several ISPs can be realized. If all networks were to meet at one exchange point, then only n connections to this exchange point would be necessary. Yet, given the large number of active ISPs and their geographical dispersion, this would still not be a feasible solution for achieving universal connectivity. Thus, for practical purposes, interconnections cannot be secured in a direct manner only. Rather, ISPs must rely on a mixture of direct and indirect interconnections to achieve universal connectivity. An indirect interconnection is established when a particular network reaches a third network by way of interconnecting with another network that is connected to this third network. Indirect interconnections can, of course, be even further removed in the sense that an indirect interconnection can also be realized by interconnecting to a network that itself is only indirectly interconnected to other networks, and so on. The interconnection agreements between the ISPs therefore determine their hierarchical relationships. There are essentially two types of interconnection relationships that ISPs enter into. The first is a customer-provider relationship, in which one ISP offers the other to deliver its traffic to and from the rest of the Internet. In such an agreement the provider of the transit service is often situated on a higher hierarchical level of the Internet than the transit taker. The second type of relationship is a peer-to-peer relationship between two ISPs that agree to exchange the traffic of their respective customers but not to transmit third-party traffic. This type of peering relationship is most often entered into by ISPs situated on the same hierarchical level of the Internet. The specific relationship between any two ISPs is subject to non-disclosure agreements generally associated with interconnection contracts. Nevertheless there are publications on the general hierarchy structure of the Internet, which interpret routing data available, for instance, from the Routeviews-Project at the University of Oregon,29 and which try to construct a picture of the relationships between the networks making up the Internet.30 From these inferences on Internet hierarchy results a general classification of ISPs in three tiers (Dierichs and Pohlmann, 2005). Tier-3, the lowest level, comprises those ISPs that are edge-networks, in the sense that they buy upstream services from other ISPs so as to have universal connectivity. They themselves do not offer transit to any third parties. The second tier comprises ISPs that have multiple connections. They offer transit to lower-level or same-level networks, have peer-to-peer relationships with some networks on their own hierarchy level, but still require an upstream provider to guarantee universal connectivity. On the highest hierarchy level are located the Tier-1 ISPs that do not need to buy transit from a third party in order to achieve universal connectivity. They have peering relationships with all other Tier-1 ISPs and offer upstream services with a guarantee for universal connectivity to lower-level ISPs. Only around ten ISPs are considered 29 See; site last visited on Feb. 15, 2008. 30 See, for instance, Subramanian et al. (2002). 45 to belong to this group of Tier-1 operators.31 The network coverage of the various interconnection agreements entered into by ISPs is explained in more detail in section 3.5, which shall also consider the implications of the interconnection contracts an ISP has for its hierarchical position in the Internet. 3.4 Design and development of the TCP/IP Protocol One important prerequisite for the development of the Internet was the programming of standards for computer communication and later for communication between different networks. This section reviews the basic principles of the Internet standards. Section 3.4.1 introduces packet-switching in general, and section 3.4.2 the TCP/IP protocol in particular. The subsequent sections 3.4.3 and 3.4.4 cover important aspects of TCP/IP communication in more detail, in particular the addressing system in IP networks, and the forwarding of data packages. Section 3.4.5 highlights newer developments in Internet logistics. Section 3.4.6 concludes by providing an outlook on ongoing Internet standardization processes. 3.4.1 Packet-switching research Traditional telephone networks use a so-called connection-oriented, or circuitswitched technology in order to link two terminal devices. With this technology, a dedicated line between caller A and receiver B is set up for the entire duration of a call. This end-to-end link is pieced together by switching through several point-topoint segments between network nodes (i.e. switches) along the way. The capacity of the link is guaranteed to this communication. In the 1960s researchers at the RAND Corporation, a think tank in California working mainly on U.S. Air Force projects, and at the National Physical Laboratory (NPL) in the UK were working on an alternative to circuit-switched communication. The motivation at RAND was to develop a secure communications system which would withstand losses of large portions of the underlying communications infrastructure, as could, for instance, be the result of a military attack.32 A circuitswitched network bears many risks in this regard because one lost central segment 31 In a recent publication the following firms were counted among the Tier-1 ISPs: MCI, AT&T, Sprint, Level3, Qwest, Cogent, Global Crossing, and Cable & Wireless (Dierichs and Pohlmann, 2005: 126). Naturally, due to mergers, acquisitions, and spin-offs, the names of the firms at the top of the Internet hierarchy have changed over time and are always subject to further change. 32 The case of nuclear war was also considered as a scenario in which large portions of the U.S. communications infrastructure could be lost. From this research at RAND the rumor got started, that the Internet was a military project with the goal of building a network resistant to nuclear attack (Baran, 1964).

Chapter Preview



Die Konvergenz der Netztechnologien, die dem Internet, der Telekommunikation und dem Kabelfernsehen zu Grunde liegen, wird die Regulierung dieser Märkte grundlegend verändern. In den sogenannten Next Generation Networks werden auch Sprache und Fernsehinhalte über die IP-Technologie des Internets transportiert. Mit den Methoden der angewandten Mikroökonomie untersucht die vorliegende Arbeit, ob eine ex-ante sektorspezifische Regulierung auf den Märkten für Internetdienste wettbewerbsökonomisch begründet ist. Im Mittelpunkt der Analyse stehen die Größen- und Verbundvorteile, die beim Aufbau von Netzinfrastrukturen entstehen, sowie die Netzexternalitäten, die im Internet eine bedeutende Rolle spielen. Die Autorin kommt zu dem Ergebnis, dass in den Kernmärkten der Internet Service Provider keine monopolistischen Engpassbereiche vorliegen, welche eine sektor-spezifische Regulierung notwendig machen würden. Der funktionsfähige Wettbewerb zwischen den ISP setzt jedoch regulierten, diskriminierungsfreien Zugang zu den verbleibenden monopolistischen Engpassbereichen im vorgelagerten Markt für lokale Netzinfrastruktur voraus. Die Untersuchung zeigt den notwendigen Regulierungsumfang in der Internet-Peripherie auf und vergleicht diesen mit der aktuellen Regulierungspraxis auf den Telekommunikationsmärkten in den Vereinigten Staaten und in Europa. Sie richtet sich sowohl an die Praxis (Netzbetreiber, Regulierer und Kartellämter) als auch an die Wissenschaft.