INNOVATION AND QUALITY OF SERVICE
The implementation of active traffic management and related traffic capacity allocations diverges from the homogeneous best-effort TCP/IP model (Knieps, 2011, p. 28; Yoo, 2013, pp.
1741, 1757). The TCP/IP stack is an integral concept of the end-to-end architecture. Within the best-effort Internet the TCP/IP stack differentiates between five complementary layers: the physical layer, the data link layer (switch-to-switch), the network layer (router-to-router), the transport layer (host-to-host), and the applications layer (process-to-process). The applications layer and the transport layer have functions executed by the host at the top of the TCP/IP stack. The introduction of QoS differentiation requires active traffic management carried out by routers and switches on the data link and network layers, thereby shifting the transport layer function from the top of the stack (the ‘edges’) to the lower network layers (the ‘core’).Contrary to much of the public debate that focuses primarily on innovations at the ‘edge’ of the Internet, innovations at higher layers and at lower layers unfold in a dynamically interdependent relationship. The capabilities supported by the network enable and constrain the types and variety of innovation possible at higher layers. At the same time, the diversity and number of innovative services at higher layers influence the value of the network and have implications for innovation and investment decisions at the network layer. Thus, in order to understand the ongoing high dynamics of the Internet, a closer look on the interplay of innovations in applications and services as well as innovations in network infrastructure and traffic services is required. As discussed with heterogeneous applications and user groups, markets for Internet traffic services can be conceptualized as platforms for Internet application services providers and end users.
The following sections explore the future role of active traffic management, which becomes even more important within the future all-IP networks.2.6.1 Active Traffic Management of Internet Traffic Service Networks
The provision of time-sensitive applications like Voice-over-IP, video conferences, and video games demands guaranteed timely and steady packet delivery. In turn, active traffic management requires a consistent incentive-compatible allocation of traffic capacities and incentive-compatible prices for network services (Knieps and Zenhausern, 2008, p. 123). Given the complementarities between lower and upper layers, innovations in application service markets may become drivers for innovations in markets for traffic services and vice versa. Innovations at the network layer include, but are not limited to, QoS differentiation in data packet transmission such as the control of delay, jitter, and packet loss.
2.6.1.1 Congestion prices and QoS differentiation
Network operators primarily make the decisions about traffic capacity investment and allocation based on the perceived business cases. Active entrepreneurial traffic management takes into account the opportunity costs of capacity usage. Within a capacity-constrained network an additional data packet may increase delay costs for all other packets. Price signals to internalize congestion externalities are well known from other fields within economics (e.g., transportation economics). For homogeneous traffic quality in the Internet they have been derived by MacKie-Mason and Varian (1995a). Different traffic classes can be provisioned relying on the differentiated services (DiffServ) architecture (Chen and Zhang, 2004, p. 370). Based on this approach it becomes possible to implement QoS differentiation by means of interclass externality pricing in IP networks. For example, in the special case of three traffic classes all data packets with the highest priority are assigned to the premium traffic class, all data packets with medium priority to the medium traffic class, and all other data packets belong to the ‘best-effort’ traffic class.
The implementation of strict priority scheduling provides a hierarchical structure to deal with congestion externalities. If the network is capacity constrained, the transmission of each additional data packet in the premium class increases delay of all lower-class data packets, thereby causing interclass externalities. Since the highest traffic class causes the highest opportunity costs for the data packets in lower traffic classes, the users of the top priority traffic class are required to pay the highest congestion fee (Knieps, 2011, p. 32).2.6.2 Vertically-related Markets
A basic characteristic of all-IP networks is the decoupling of application services and traffic management such that service-related functions are independent from underlying traffic-related technologies. Since all transmissions are IP-based there may be a combination of interconnected broadband infrastructures involved. Although application services and traffic management services are technically separated, vertical relations between application service providers and traffic service providers may exist. The question arises whether application service providers owned or tightly cooperating with a traffic service provider or paying for QoS will be granted higher traffic priority (‘access tiering’). If so, do such arrangements constitute forms of unfair competition and/or a violation of the vision of a ‘neutral’ and ‘open’ Internet?
During the past decades the IETF has evolved as a forum for developing standards for basic building blocks (e.g., IntServ/RSVP, DiffServ, combination versus integration of DiffServ and IntServ) as a basis for flexible open traffic architectures that enable the provision of a variety of traffic qualities within multi-purpose traffic networks. Instead of standardization of a specific QoS differentiation architecture, the IETF only proposes configuration guidelines for DiffServ classes, based on the standardized building blocks. This provides freedom to entrepreneurs to implement a specific QoS traffic architecture derived from the applications’ quality requirements.
The open set of flexible multipurpose QoS architectures capable of supporting a variety of traffic qualities required for different application services can be considered as a ‘meta-architecture’, termed Generalized DiffServ architecture (Knieps, 2015b, p. 739).The entrepreneurial selection of a specific QoS differentiated architecture as an implementation of the Generalized DiffServ architecture provides a flexible common multi-purpose traffic architecture that can support time and non-time-sensitive applications and services. QoS differentiated pricing, taking into account the opportunity costs of traffic capacities, is incentive compatible because discrimination between application services on the basic of traffic capacity requirements is avoided. The general principle of rivalry for network resources for different traffic classes is applied. Under QoS-differentiated pricing the minimal traffic quality of the lowest traffic class results endogenously. It differs from the best-effort average traffic quality of TCP. Consequently, an artificial market split between best-effort TCP and specialized services within Generalized DiffServ architecture would not be incentive compatible and unstable. Within the Generalized DiffServ architecture the use of best-effort TCP with passive traffic management at the edges (hosts) conflicts with the active traffic management required for the provision of heterogeneous traffic qualities (Knieps, 2015b).
2.6.3 Network Neutrality
Openness has long been seen as one of the key drivers of Internet-based innovation. With the migration to all-IP networks and the differentiation of applications, services and demand, the traditional best-effort approach to traffic management has come under considerable strain. A long debate exists on whether the openness of the Internet needs safeguarding by means of regulatory interventions into active traffic management (Schwartz and Weiser, 2009; Sidak and Teece, 2010; Schuett, 2010; Kramer et al., 2013).
Conducted under the label of network neutrality, arguments usually point to the ambiguous incentives of ISPs for whom some applications are complementary while others compete with their own services. Economic concerns related to the implications of quality of service differentiation on innovation and investment and political concerns related to the freedom of speech are articulated and often poorly separated (Bauer and Obar, 2014). Few specific cases of discrimination are known, with the BitTorrent case in 2008 the most visible instance, revealing that Comcast had engaged in practices to throttle certain types of traffic (FCC, 2008). The case illustrated the problems caused by overuse by some user groups and by intransparent rationing by network operators (Wu, 2003). In the USA, two issues drove the debate. First, there was a concern that unregulated market power in broadband access networks would be shifted to the markets for Internet access services and subsequently abused to exploit Internet application service providers (Economides, 2008, p. 210). At the same time there were concerns that quality differentiation in the Internet would undermine its vibrant, ‘permission-free’ innovation system. While such concerns are not entirely invalid, the major arguments against network neutrality regulation have been that general competition law and consumer protection laws apply to competition in the Internet. Instead of prohibiting price and quality differentiation by ISPs, it seems a more appropriate strategy to regulate market power at its roots, that is, where monopolistic bottlenecks in broadband infrastructures prevail (Knieps and Zenhausern, 2008, p. 127).A related question is whether it would be beneficial from a welfare point of view to allow QoS differentiation but require the provision of a minimal quality to restrict quality of service limitations imposed by other broadband providers. A crucial advantage of minimum quality regulation is that in contrast to network neutrality requirements it is consistent with charging content providers for carriage and that it does not preclude transmission qualities higher than the minimum quality, although presumably supplied at higher prices.
A minimum quality standard, however, would not be without potential drawbacks, among others bearing the risk of suppressing some competition through quality differentiation. Compared to enforced uniformity through net neutrality rules the cost of minimum standards would appear far smaller (Brennan, 2011, p. 69). From the perspective of all-IP Internet, guaranteed high traffic quality becomes crucial for highly time-sensitive applications. Other applications only need low priority transmission. If regulators would prescribe a minimum traffic quality the freedom of Internet entrepreneurs to choose traffic service classes would be hampered and a demand for low traffic quality might go unfulfilled (Knieps and Stocker, 2015, p. 51). The entrepreneurial flexibility to individually develop tailored QoS traffic architectures may result in a large variety of QoS differentiation strategies. In a fast-paced technological environment, preserving this evolutionary search for improving QoS differentiation depends on the flexibility of entrepreneurs envisioned in the implementation of the Generalized DiffServ architectures. Regulatory interventions will likely impede this type of discovery process.One important argument within the network neutrality debate has been that the besteffort principle of the TCP/IP - which implies a zero-price rule for transactions between ISPs and content providers and the recovery of network costs from end users - promotes content and application service innovation. There is considerable evidence for such a process but it is not the only type of innovation relevant in the Internet ecosystem. A ban on charging termination fees to content providers would effectively subsidize the creation of new application services. The prohibition of fees facilitates entry of new application service providers and also empowers consumers to participate via user-generated applications and content. Due to the trade-off between possible welfare gains if content production is increased and potentially higher access or usage fees to consumers, the final answer to the highly controversial question of network neutrality regulation remains open (Lee and Wu, 2009, p. 66). As outcomes are highly contingent on specific modeling assumptions no clear-cut conclusion about the relationship between net neutrality regulation and the innovation incentives of Internet traffic service providers or content providers has emerged (Choi and Kim, 2010).
The conclusion from the controversial literature on network neutrality regulation seems to be that a regulatory implementation of a zero-price rule is not well justified. The opposite may be the case because of the potential importance of ‘innovational complementarities’ (Bresnahan and Trajtenberg, 1995, p. 84). Since the Generalized DiffServ architecture has the characteristics of flexible multi-purpose technologies the (derived) demand for specific traffic qualities to enable new application services can be successfully matched, which strongly stimulates the entrepreneurial search of innovative application services and related application software. Moreover, innovations in Internet traffic architectures that facilitate more differentiated transmission qualities can also cause the search for new application services. Consequently, the innovation potential for application services and Internet traffic are interrelated. A key challenge for the policy framework is therefore to balance these two interdependent innovation dynamics, one facilitated by non-discriminatory access to network and the other dependent on differentiation.
2.6.4 The Entrepreneurial Search for Active Traffic Management Architectures
The interactive dynamics of complementary general (multi-) purpose technologies and applications within the Internet can build on the GPT literature. Although this research has focused on the role of key technologies on aggregate economic growth, its overall framing of ‘innovational complementarities’ is relevant for understanding the dynamics of the Internet. According to Bresnahan and Trajtenberg (1995, p. 83) innovation in the upstream GPT increases the productivity of R&D in downstream application markets. In turn, innovations and improvements in applications raise the return to further advances in the GPT.
This perspective re-emphasizes the analysis provided above that the major driver of innovations within the Internet is not limited to applications (Simcoe, 2012). Innovation may also be stimulated by positive feedback effects from the all-IP infrastructure and Generalized DiffServ architecture, which function as GPTs for applications and services. Therefore, it is important that the GPTs, both on the broadband infrastructure level as well as on the traffic architecture level, remain open for innovative evolution, taking into account requirements of the application side. Although complementarities between different GPTs as well as between GPTs and application services are important, the entrepreneurial search process on the specific markets for all-IP infrastructure capacities, for Internet traffic provision (combining infrastructure capacities with Internet logistics) as well as Internet applications (search engines, content, PC software etc.) is also vital.
2.7