INDUSTRIAL ORGANIZATION OF THE DYNAMIC INTERNET
2.3.1 Technological Basics and their Dynamics
During the initial phase of Internet evolution it was meaningful from an economic perspective to distinguish between Internet periphery and service provision (Knieps, 2003, p.
218). Elements that were viable on their own were considered as part of the periphery (e.g., terminal equipment, content, local telecommunications network infrastructure, and long-distance telecommunications infrastructure). In contrast, Internet traffic services (based on telecommunication infrastructure capacities and Internet logistics) as well as Internet application services (based on Internet traffic) could be considered to form the nucleus of the Internet. The convergence towards all-IP networks has rendered the distinction between telecommunication infrastructure and the Internet untenable as both are seamlessly integrated. Instead all-IP network infrastructures are better conceptualized as GPTs.These developments point to the need for a more general approach towards the evolution of Internet architectures: the choice between where to place functionality and ‘intelligence’, on the edges, in the network core, or a hybrid. The optimal configuration depends on the relative costs of different solutions, path dependency of technological decisions, alternative entrepreneurial business models and so on. As broadband infrastructures are becoming part of an all-IP network in which different levels of service quality are differentiated to properly support services such as Voice-over-IP (VoIP), the traditional specialized high-quality networks (e.g., PSTN [Publicly Switched Telephone Network], ISDN [Integrated Services Digital Network]) are phased out. In this all-IP environment an industrial organization of the Internet can build on a disaggregated representation distinguishing all-IP broadband infrastructures, markets for Internet traffic and markets for application services.
Broadband access networks use different types of fixed network technology, including copper wire, coaxial cable, and fiber. Moreover, several mobile access network technologies are available and increasingly integrated with fixed networks into seamless access infrastructures (ITU, 2014; OECD, 2014; Tripathi and Reed, 2014). The specific technology mix of access networks varies and is to some degree path dependent on the upgrading strategies chosen by the network operators (Knieps and Zenhausern, 2015). Various generations of DSL technologies configure broadband connections based on copper wire. Innovations such as vectoring have expanded the capacity of copper access loops far beyond initial constraints. Cable systems also were able to increase data rates by improving the protocols used for data transportation, most recently DOCSIS 3.1 (Data Over Cable Service Interface Specification), which allows access speeds up to the gigabit range. Presently, different fiber optical network solutions including fiber to the curb (FTTC) and fiber to the home (FTTH) allow the highest data rates.
These fixed network technologies are complemented by wireless access technologies, which became available in the 1990s and offered new options to connect at the edges of the network. Wireless local area networks (WLANs, particularly Wi-Fi) and wireless personal area networks (WPANs, particularly Bluetooth) facilitate seamless connectivity at different scales. Mobile Internet access became available using 2G (e.g., GSM, CDMA) wireless technology although data throughput was low. With the advent of 3G (e.g., UMTS), 4G wireless networks (e.g., LTE [long-term evolution] and WiMax [Worldwide Interoperability for Microwave Access]) and presently 5G platforms, data rates in the 100 Mbps range can be supported and have greatly expanded the uses of mobile broadband, including live video streaming, mobile games, and other data-intensive applications.
All-IP access infrastructures increasingly combine these component technologies.
Because the alternative network platforms have different costs, this allows configuring a least-cost network infrastructure by rolling out an appropriate mix of technologies. A growing number of network operators take advantage of this flexibility by integrating various fixed and wireless network elements into a seamless access infrastructure. Upgrading costs of DSL acceleration technologies like vectoring are much lower than investment costs to establish FTTH (Zhao et al., 2014, p. 11). Capacity and speed over cable networks depends on the type of upgrading investments to provide bidirectional communication, often achieved by replacing coaxial parts of the network with fiber plant. Fixed access technologies are increasingly complemented by wireless platforms with a growing number of users opting for wireless only access. The development of access networks does not seem to converge toward a dominant platform, due to heterogeneous irreversible investments in alternative access technologies and heterogeneous consumer demands. Rather, the pervasive use of IP-based transmission supports the coexistence of a variety of access network infrastructures that are integrated by the IP layer (Knieps and Zenhausern, 2015).2.3.2 The Dynamics of Internet Traffic Management
Given the nature of the Internet as a network of networks, interconnection among the proliferating number of ISPs is critical (Noam, 2001). Interconnection has evolved considerably from the initial arrangements. Earlier public peering via network access points (NAPs) was often plagued by congestion and delay due to rapidly growing traffic volumes. Consequently, private peering arrangements evolved soon after the transition to a commercial Internet. In the mid-1990s two basic forms of incentive-compatible contracting between different Internet traffic providers dominated: transit and peering. Under transit, a traffic provider is selling data packet transmission to its customers to and from any Internet destination.
Under peering, bilateral or multilateral arrangements between traffic providers on a barter trade (bill and keep) basis without payments are provided. With the increasing asymmetries and heterogeneity of traffic flows, additional arrangements such as paid peering have emerged (see also Clark, Lehr and Bauer, Chapter 16 in this volume). Only data packets between the customers of peering traffic providers (as well as data packets of transit customers of their customers) are transmitted (European Commission, 1998, p. 7). Data packets to and from customers of non-peering networks that are served by other peering partners are not transmitted.2.3.2.1 The traditional universal connectivity paradigm
The pursuit of universal Internet connectivity has resulted in a hierarchical network structure with peering and transit arrangements (e.g., European Commission, 1998, p. L116/6; Laffont and Tirole, 2000, p. 268; Besen et al., 2001; Cremer et al., 2000, p. 443). Consequently, interconnection of networks in various institutional arrangements has been a major driver of the traditional Internet. Internet exchange points (IXPs) emerged as central switching platforms that allow networks to interconnect directly (Vanberg, 2009, p. 81). Initially, a set of typically larger traffic providers had peering arrangements with each other and enjoyed universal connectivity without the necessity of transit arrangements (called tier 1 or top-level traffic providers).
ISPs not belonging to the ‘club’ of tier 1 networks could benefit from their universal connectivity by buying transit from one of them. Within the traditional Internet, top-level providers were considered to compete in an upstream (‘backbone’) market distinct from the downstream market for Internet access (Besen et al., 2001, p. 292). If traffic is symmetric between some traffic service providers and asymmetric between others, combinations of peering and transit arrangements may be chosen. Secondary peering traffic providers offer their customers a mix of their own peering-based connectivity combined with transit to and from the remaining destinations.
Incentives may arise for non-tier 1 providers to peer with another non-tier 1 provider to exchange data packets between these two networks and thereby bypassing the transit options provided by the tier 1 providers. Indeed, during the past decade the historical hierarchical arrangements have started to change with an increasing number of peering arrangements among tier 2 and even tier 3 networks. This has considerably reduced the role and power of tier 1 networks (claffy and Clark, 2014).2.3.2.2 The evolution of partial transit and paid peering
The transition from narrowband to broadband has gone hand in hand with a diversification of services and a reconfiguration of the architecture of hosting and service provision. As a result, traffic flows among networks have become increasingly asymmetrical. Thus incentives arise to develop more flexible contract arrangements between traffic providers dealing with heterogeneous traffic demand (Faratin et al., 2008, p. 58). Paid peering increases the flexibility of peering arrangements if asymmetric traffic flows between two potential peering partners are to be compensated by side payments. Partial transit limits universal connectivity by restricting the scope of sender and/or receiver addresses and thereby reducing the opportunity costs of transit obligations.
2.3.2.3 Overlay networks
Overlay networks are designed to enhance or modify the basic functions of traffic handling on top of the traditional best-effort TCP/IP Internet architecture. They include routing overlay networks, CDNs, P2P, and security overlay networks. Overlay networks may be commercial, cooperative, or peer-to-peer based. Whereas commercial overlay network providers (e.g., Akamai) create successful business models, a large scope for noncommercial activities evolves where peers are both suppliers and consumers (prosumers). CDNs consist of caches of content copies distributed across the Internet in order to economize response time and server loads and to reduce bandwidth costs.
The purpose of routing overlay networks is to reduce routing delays of the best-effort TCP/IP Internet by running the overlay protocol software. Although a routing overlay network cannot change the TCP routing procedure, it can optimize the sequence of overlay nodes that a data packet traverses to its destination. A popular example of a peer-to-peer overlay network is file sharing among equally privileged peers acting as prosumers (Clark et al., 2006, p. 3). Overlay networks can to some extent fulfill complementary functions of traffic management not provided within the TCP/IP Internet. However, overlay networks cannot substitute for active traffic management within all-IP networks.2.3.2.4 The challenge for active traffic management
In the following subsection only a short preview for active traffic management focusing on negative externalities due to congestion is provided whereas a further analysis of QoSbased traffic management is provided in section 2.6.1. Different definitions of congestion are applied by network engineers, which differ from an economic definition of congestion (Bauer et al., 2009, p. 8). Queuing theory considers traffic as congested if within a time interval the arrival rate into a system exceeds the service rate. Alternatively, a network operator may define congestion as the (average) load on a network over a particular period of time exceeding a specified level. Within TCP congestion control the focus is on packet dropping and the impact of send-rate reduction on queuing. In contrast an economic definition of congestion is based on the welfare economic concept of negative externalities, already analyzed by Pigou (1920). When an increase in the use of a facility imposes a cost (negative externality) on the existing users the facility is considered to be congested.
This economically well-founded concept is widely accepted in network economics and can be applied to any network architecture (e.g., Knieps, 2015a, Chapter 3). In the context of packet data transmission it has been analyzed by MacKie-Mason and Varian (1995b). A congestion charge can be assessed to avoid overusage of a facility, thus mitigating congestion externalities. The socially optimal congestion fee reflects congestion cost. Since TCP controls each traffic flow individually, a differentiation between heavy and light users of capacity is not possible. TCP allocates a higher share of capacity to intense users and a lower share to light users. Moreover, TCP cannot provide prioritization of data packets and quality of service guarantees (Knieps, 2011, p. 27). As long as active traffic management with price and quality differentiation is not implemented, network usage restrictions aiming to limit capacity consumption of heavy users are to be expected. Moreover, network operators may pursue a bifurcation of the network into a public and a private Internet. It is doubtful whether, within all-IP networks, such a market split can be a stable configuration in the long run (Knieps and Stocker, 2015).
2.4