You are on page 1of 187

WIK-Consult • Final Report

Study for the European Commission

The Future of IP Interconnection:
Technical, Economic, and Public Policy Aspects

Authors: J. Scott Marcus, Dieter Elixmann Contributing Author: Kenneth R. Carter Senior Experts: Scott Bradner, Klaus Hackbarth, Bruno Jullien, Gabriele Kulenkampff, Karl-Heinz Neumann, Antonio Portilla, Patrick Rey, and Ingo Vogelsang WIK-Consult GmbH Rhöndorferstr. 68 53604 Bad Honnef Germany

The opinions expressed in this study are those of the authors and do not necessarily reflect the views of the European Commission.

Bad Honnef, 29 January 2008

Final Report: The Future of IP Interconnection

I

Contents
Figures Tables Key Concept Boxes Executive Summary 1 Introduction 1.1 The changing electronic communications environment 1.2 The economics of Interconnection today 1.2.1 The distinction between access and interconnection 1.2.2 Current European arrangements for switched network interconnection 1.2.3 Negotiated arrangements 1.2.4 The GSM Association’s IP-based Interconnection model 1.3 Emerging challenges to IP network interconnection 1.3.1 Decoupling of the network from the service 1.3.2 Migration to NGN 1.3.3 The evolution of the access network 1.3.4 Changing cost structure 1.4 Implications of interconnection payments for consumer and supplier welfare 1.5 Regulatory objectives 1.6 Methodology 1.6.1 Desk research 1.6.2 Stakeholder interviews 1.6.3 Consolidation and analysis of results, formulation of recommendations 1.7 Structure of this report 2 Underlying drivers of change 2.1 Technological and market evolution 2.1.1 De-coupling of the service from the network 2.1.2 Continued price/performance improvements due to Moore’s Law and to enhancements in fibre optic transmission (e.g. DWDM) and optical switching and cross-connection 2.1.3 Fundamental changes in the delivery of voice and video services 22 24 V V VI VII 1 2 3 3 4 5 7 8 8 11 12 12 13 14 16 16 16 17 17 18 19 19

1.3 Where management of QoS is likely to be beneficial 27 29 29 31 32 33 33 35 39 40 43 43 43 45 47 48 49 49 52 63 69 71 72 76 77 77 80 80 82 84 .1 Transit 3.3 Differentiated Quality of Service (QoS) in the Internet today 4 Differentiated Quality of Service (QoS) 4.3 Implications 3.2 The impact of management of QoS 4.1 Key underlying economic principles 3. Peer-to-Peer (P2P).2.3.2 Wholesale arrangements among switched fixed (PSTN) and mobile (PLMN) telephone networks 3.4 Implications for IP-based traffic patterns 3 Economic theory and practice: Interconnection today and tomorrow 3.3 IP interconnection models 3.5 Future evolution of higher-level routing functions 2.1 Web 2.1.6 Gradual deployment of IP version 6 (IPv6) 2.2 The technology of QoS management 4.1 Application requirements 4.2.1 The performance of IP-based networks in the absence of QoS management 4.2 Market power 3.1.1.4 Two-sided markets 3.II Final Report: The Future of IP Interconnection 2.2 Peering 3.1 Externalities 3.2.3 Long term versus short term effects 2.2.4 Gradual modernization of exterior IP routing 2.1.2 Traditional telephony interconnection models 3.1.2. and the shift to user-provided content 2.1 Retail arrangements for voice telephony 3.2.1.0.3.1.7 Changes in network back-haul 2.8 Replacement of the circuit-switched access network with a high speed IPbased network 2.2 Broader changes in the use of electronic communications 2.1.3 Transaction costs 3.3.2.2.2 The changing environment within the home 2.

3.2.2 IMS and NGN – standards and implementation 5.1 Interconnection of VoIP 5.4 Technologies used to manage QoS 4.1.1 Quality differentiation. innovation.1. and market power 4.3 Fixed.3 Implications for European policymakers 5 NGNs.5 Slow deployment between service providers to date 4.4 Implications of the different migration paths 85 89 89 89 90 90 91 92 94 95 95 97 98 98 99 99 101 114 116 117 122 122 122 123 124 124 125 125 125 126 .2 Incumbent or competitive entrant? 5.Final Report: The Future of IP Interconnection III 4.2 Technical characteristics of NGN 5.3.3.4 Incentives for interconnection 4.1 Service differentiation 4. or historically IP-based? 5.2 Prospects for third party services over NGN 5.4 The Network Neutrality debate in the United States and elsewhere 4.4. social welfare.3.4 Different market players – different evolutionary paths to NGN 5.3 NGN/IMS technology and interconnection at the service level and the IP level 5.3 Transaction costs 4.4.4.2 Vertical integration and other vertical relationships between network operators and applications and content providers 4. and IP interconnection 5.4. mobile or both? 5.1.1. Voice over IP (VoIP).1 Interconnection at various levels 5.2.1 Rationale for VoIP peering 5.4.3.3 The economics of differentiated QoS 4.2.1 Implications of standardized NGN equipment 5.1 Historically a telco.3.3 ENUM and VoIP peering 5.3.4 VoIP peering and market power 5.3.3 Implications for Carrier Selection and Carrier Pre-Selection 5.4 Related threats to competition 5.2 Number portability and VoIP peering 5.4.2 Network externalities 4.4.3.

5.4 Interconnection payments – voice call termination 7.6 Technical considerations in the evolution of the IP protocol References Annex 1: Annex 2: Survey of recent published work on IP interconnection Glossary of Terms 129 130 130 131 132 134 134 136 137 138 139 139 140 140 142 145 146 148 153 162 .4.5.2 Number of Points of Interconnection (PoI) 126 126 127 5.6 Technical considerations in the evolution of the IP protocol 7 Recommendations 7.4.1 The level of charges for IP data traffic 6.2 The level of charges for voice 6.1 The evolution of the market 6.1 Traditional remedies versus new remedies 5.3 Interconnection obligations – voice call termination 7.4 The level of charges among IP-based network operators 6.4 A risk of arbitrage? 6 Conclusions 6.5.2 The call termination monopoly 6.1 Interconnection obligations – IP traffic 7.5 The migration to an IP-based NGN 6.5 Network neutrality 7.5 Regulation and the transition period to VoIP and NGN 5.3 Interconnection obligations for data 6.3 The changing average incremental cost of service during the transition period 128 5.2 Wholesale payments for IP traffic interconnection 7.5.IV Final Report: The Future of IP Interconnection 5.

Fixed-to-mobile interconnection fees in the EU25. 2001-2004 Figure 16. 2004-2006.S. 2000-2004 Figure 15: Fixed-to-mobile interconnection fees in the EU15. Figure 17: Mobile revenue per MoU versus MoUs per month Figure 18: Two peers and their respective transit customers Figure 19: Application requirements for stringent QoS Figure 20: MIT reference model for inter-provider QoS Figure 21: End-to-end SLAs and charging with the GSM-A IPX Figure 22: Different Forms of VoIP Implementation Figure 23: The ITU’s layered NGN architectural model Figure 24: Layered view of the IMS Model Figure 25: Functional elements of IMS Figure 26: Fixed and mobile interoperability under ETSI TISPAN Figure 27: Evolutionary paths to IMS NGN 9 10 19 21 22 24 37 38 39 42 44 48 53 56 56 57 65 73 79 88 109 111 115 118 119 120 121 Tables Table 1: Coasian arrangements: Backbone peering versus U. Bill and Keep 75 .Final Report: The Future of IP Interconnection V Figures Figure 1: The TCP/IP Reference Model Figure 2: IP-based applications running in your Personal Computer (PC) Figure 3: The NGN layered structure Figure 4: Convergence of content delivery Figure 5: Introduction of Intel Chipsets 1992 -2006 Figure 6: The value chain for the provision of VoIP Figure 7: Protocol hierarchy of E-Home related standards Figure 8: Timeline of E-home related standards Figure 9: E-home value chain and potential business models Figure 10: Cisco estimates of global consumer Internet traffic Figure 11: Network Externalities Figure 12: Two-sided markets Figure 13: Calling Party's Network Pays (CPNP) wholesale arrangements Figure 14: Fixed-to-fixed interconnection fees in the EU15.

Interconnection Key Concept Box 3: Call termination fees and the migration to IP Key Concept Box 4: Call termination fees: Mobile networks Key Concept Box 5: Migration to Next Generation Networks: Regulatory Objectives Key Concept Box 6: Technological and market evolution Key Concept Box 7: Symmetric vs. cont’d Key Concept Box 16: Carrier ENUM and Public ENUM Key Concept Box 17: Mobile Interconnection Key Concept Box 18: Mobile IP Interconnection Key Concept Box 19: VoIP Interconnection Key Concept Box 20: Internet vs. Peer-to-Peer (P2P).VI Final Report: The Future of IP Interconnection Key Concept Boxes Key Concept Box 1: Note to the reader Key Concept Box 2: Access vs. and the shift to user-provided content Key Concept Box 9: Internet Traffic Growth Key Concept Box 10: Reciprocity vs.0. NGN Key Concept Box 21: Conclusions Key Concept Box 22: Conclusions cont’d 1 4 6 14 15 18 34 35 41 61 79 94 96 97 100 100 105 108 112 117 130 133 . Symmetry Key Concept Box 11: Quality of Service economics and engineering Key Concept Box 12: Network Neutrality Key Concept Box 13: Net Neutrality: US vs. Europe Key Concept Box 14: VoIP peering Key Concept Box 15: VoIP Peering Issues. asymmetric relationships Key Concept Box 8: Web 2.

• • This Executive Summary follows roughly the same sequence of presentation as the report itself. these networks are physically and logically converging to IP-based Next Generation Networks (NGNs). the drivers of change. but also in terms of associated regulatory obligations. the economic theory of differentiated Quality of Service (QoS). Access. With the increasing importance of the Internet in the global economy. the significance of IP-based interconnection will become even greater in the years to come. Notably. is out of scope for this report. not only at a technical level. Consistent with the general philosophy of the European framework. We attempt to clearly delineate between the two. we have attempted to draw clear distinctions in a number of areas that are sometimes blurred together in the relevant literature. In the following sections. and has for the most part been highly effective. We see signs that mobile networks are already following a somewhat different evolutionary path than fixed networks. Retail versus wholesale level: There are linkages between the retail and the wholesale levels. Indeed. Interconnection arrangements for switched PSTN/PLMN networks have been markedly different than those for IP networks. A number of technological and market developments pose challenges to IP-based interconnection. the economic theory of interconnection as it relates to the Internet and to the PSTN/PLMN. we have taken the position that regulatory remedies should be imposed only where necessary. The convergence of these networks raises difficult questions as to how interconnection should be regulated going forward. with potentially large regulatory implications. Our conclusions and recommendations reflect these differences. and our recommendations to the European Commission. Fixed versus mobile networks: Previous papers on IP-based NGN interconnection have often ignored the differences between fixed and mobile networks. but they are not the same thing. we have taken the position that needed remedies should if possible be imposed only on the wholesale level. moreover. In the study. not on the retail level. and to traditional interconnection in the fixed (PSTN) and mobile (PLMN) networks as well. it is the convergence to NGNs that makes this study particularly timely.Final Report: The Future of IP Interconnection VII Executive Summary Network interconnection by means of the Internet Protocol (IP) has been a vital enabler of the Internet’s ubiquity and success. . Among them: • Access versus interconnection: Our focus is on interconnection between independent networks. we discuss the goals and the challenges of regulation. whereby one network operator uses facilities of another in support of the former’s own services and customers. IP-based interconnection has usually been achieved without explicit regulatory obligations.

Regulation also has a role to play in ensuring that necessary services are deployed even in cases where market mechanisms would not suffice. lawful intercept (wiretapping). 1 See the Framework Directive. thus changing the character of competition and possibly introducing new challenges to competition. Examples that are relevant to NGN. and access by those with disabilities. price. fixed telecommunications and cable television) to offer equivalent services. This evolution makes it possible for different underlying platforms (for example. include access to emergency services (112). comparable to. Drivers of change The most noteworthy driver of change is the convergence of the network. This development potentially represents a potential boost to competition.1 As the fixed and mobile voice networks migrate to IP. and quality of electronic communication services. Article 8. and may be motivated to try to thwart the emergence of these new forms of competition if they can. and that are already present in today’s Voice over IP (VoIP) services. and to avoid barriers to innovation and to efficient investment in infrastructure. . posing as it does risks of distortion of competition. specifically in regard to IP interconnection and also generally. In particular. IP-based converged Next Generation Networks be less than. inasmuch as some existing operators may perceive this evolution as a competitive threat. to ensure that all users derive maximum benefit in terms of choice. This same evolution enables bundled offers of multiple services to the end-user. As a closely related phenomenon. This evolution represents both an opportunity and a challenge. Independent service providers can compete with established network operators to offer data. voice and video services to end-users. market power has been a central concern of regulation. voice and video.VIII Final Report: The Future of IP Interconnection Regulatory challenges. convergence implies that the service becomes largely independent of the network. or greater than market power in current voice networks? The answer is not yet altogether clear. Will market power in the emerging. regulatory goals Regulators seek. with a single integrated IP-based network (whether called an NGN or not) delivering some combination of data. potentially benefiting competition. regulation will need to be carefully reviewed with these objectives in mind. to minimize any distortion or restriction of competition in the electronic communications sector. and a great step forward for consumer choice.

These changes signal shifts in traffic volumes and in traffic patterns. so they are generally subject to regulation in the European Union today. video. but the rate of growth (contrary to what many have assumed) is gradually slowing in percentage terms over time. Voice over IP (VoIP) usage will continue to grow. through improvements in Dense Wave Division Multiplexing [DWDM]). Any combination of voice. or mobile infrastructure (albeit possibly with differences in price and performance). and price/performance continues to improve dramatically over time. . Technology continues to evolve. one ISP agrees to carry the traffic of a customer (possibly also an ISP) to third parties. 2 Other IP interconnection arrangements exist. Economic theory: Interconnection in the switched network and in the Internet IP-based interconnection is normally implemented by a mixture of peering and transit. such as mutual transit. telecommunications. but also for communications bandwidth (for instance. generally for a fee. two Internet Service Providers (ISPs) agree to exchange traffic solely among their respective customers. and the associated shift from pure client-server applications to peer-to-peer (P2P) models. but they are less frequently used. Peerto-peer traffic will continue to grow. Video over the Internet is likely to represent a major driver of future traffic growth. and data can be delivered over cable television. not only for computational power and memory (Moore’s Law). with a current global average of 50-60% growth per year and a European average perhaps slightly higher. with transit.Final Report: The Future of IP Interconnection IX This same phenomenon of IP-based convergence implies that services are no longer tied to a single physical transmission platform. In most of the world. but will never represent a large fraction of total Internet traffic. and do not (in most cases) depend on any regulatory obligations. where the network of the party that places (originates) a phone call makes a wholesale payment to the network of the party that receives (terminates) the call. sometimes without payment. telephony service providers adhere to wholesale payment arrangements known as Calling Party’s Network Pays (CPNP). Economic theory shows that network operators (even small network operators) would prefer to set these fees at exceptionally high levels. IP-based traffic continues to grow dramatically.2 With peering. Another major driver of change is the evolution to user-provided content. but will decline somewhat as a fraction of all Internet traffic. These freely negotiated arrangements result in a richly interconnected Internet.

and is comparable to that of firms that subsidize the sale of razors or computer printers in the hope of “locking in” longer term highly profitable sales of razor blades or ink. however. Under these alternative arrangements. and may in some cases depress use below the level of efficient monopoly price for the operators as well. • • CPNP systems with high mobile termination fees may tend to drive faster take-up of mobile services. or some third model (possibly a blend of the two)? To answer this. respectively. Mobile telephone services in a CPP/CPNP system tend to have some combination of low (or zero) monthly fees. it is clear that current CPNP arrangements already have a substantial negative effect on welfare. Carrying them forward at current fee levels into the realm of IP-based NGNs would lead to additional problems. The inevitable question is. Canada. IP-based NGNs represent the convergence of the Internet and the traditional switched PSTN/PLMN network. Service providers are willing to subsidize these fixed prices because usage prices are greatly in excess of average incremental cost. the network operators freely negotiate termination fees. The high retail prices depress use of the service to levels far below those that are efficient. even when the fees are regulated. for a variety of reasons: 3 In effect. . subject to obligations to interconnect and typically subject to the requirement that both networks that are parties to the same agreement must be subject to the same per-minute fees.4 We conclude that CPNP arrangements are already problematic today. with minor exceptions. especially for the mobile network.S. in a number of respects: • They tend to lead to inefficiently high wholesale termination fees.. and heavily subsidized handset prices. High wholesale mobile termination fees in effect create collusory incentives to maintain high per-minute (mobile) retail prices.3 which can be positive in developing countries. further stimulus to mobile adoption for the EU27 is not needed. 4 It is sometimes also argued that these high retail prices inhibit undesired calls (the telephony equivalent of SPAM). the switched network model. it is helpful to begin with an assessment of how the two sets of interconnection arrangements operate in their current milieus. and Singapore) use alternative arrangements for mobile operators and for non-dominant fixed operators (while still requiring dominant fixed operators to use CPNP). For reasons explained more fully in the report. They often choose to set these fees to zero (a system known as Bill and Keep). users of fixed services subsidize mobile take-up. Given that they also inhibit desired calls.X Final Report: The Future of IP Interconnection A few countries (the U. low initial fees. we find the argument unpersuasive. This behaviour is rational for the firms. should interconnection of future NGNs be based on the Internet economic model.

If termination fees were to remain at current levels. Minutes of voice use will bear even less relation to the cost of the network going forward than they do today. or a modernized version of them. they will invite arbitrage that seeks to correct for the irrationalities in the pricing system. preferably zero. the payment makes no sense. The migration to IPbased NGNs might possibly compound this familiar termination monopoly with other forms of market power. Alternatively. Notably. and that his or her network should consequently bear the total wholesale cost of the call. the complexity of the system going forward leads us to feel that negotiated arrangements are to be preferred to regulated arrangements wherever they are likely to lead to efficient outcomes (and not to prices that reflect market power). Voice traffic represents a small and declining fraction of the cost of an NGN. the migration to IP-based NGNs will not ameliorate the termination monopoly (the tendency to charge high prices for termination) until and unless mechanisms emerge to enable more than one telephone service provider to terminate calls to a single telephone number. but it is associated only with the voice service. • • • All of these considerations lead us to feel that interconnection fees should be much lower than they are today. The rationality of this assumption is dubious in the current network.Final Report: The Future of IP Interconnection XI • The call termination fee is intended to capture costs of the network. To the extent that wholesale termination fees are unrelated to cost. If the voice service provider is not the same company as the network provider (as can readily be the case in an NGN). The payment assumes that the party that placed the call is the sole beneficiary of the call. These concerns suggest that waiting for the migration to IPbased NGNs to implement changes might be a self-defeating strategy. and even more so in an IP-based world where the direction of the call can trivially be reversed. the disparity between mobile termination rates and fixed should be vastly less than it is today. At the same time. As long as Europe operates under CPP/CPNP retail and wholesale arrangements. rather than enabling IP-based interconnection arrangements to evolve in a healthy and natural way. . they might choose an IP interconnection strategy that attempts to lock in the current inefficient arrangements. that he or she should bear the total retail costs of the call. They might perceive the migration as an unacceptable regulatory risk. we anticipate that many mobile operators and some fixed operators might choose not to evolve their networks to IPbased interconnection.

Over the past two decades. and the likelihood of packet loss) for perhaps a decade. 5 For one-way “streamed” audio or video.XII Final Report: The Future of IP Interconnection Economic theory: Quality of Service (QoS) in the switched network and in the Internet The Internet was designed with non-delay-sensitive applications such as email in mind. • • Differentiated QoS has been a major theme for network operators migrating to NGN. there has been growing interest in enhancing IP-based networks to provide greater assurance that traffic is consistently delivered quickly enough. Reasons for slow deployment include: • Inability of most customers to perceive much of a quality difference most of the time. but rather to ensure that consumers are empowered to make informed decisions. we do not see either promoting or inhibiting the deployment of differentiated QoS as a proper matter for industrial policy. in the absence of market power. As of now. differentiated QoS in IP-based networks is most often implemented using a combination of DiffServ and MPLS technologies. we note that. but rarely between networks. We think that the market should determine whether differentiated QoS deploys or not – to a first order. and The transaction costs associated with negotiating QoS-sensitive interconnection arrangements with a large number of interconnection partners. 6 Today. Consequent lack of customer willingness to pay much of a premium for better service. Ten years ago. it seems to us that the degree of market acceptance going forward of differentiated QoS between providers remains unclear. even for delay-sensitive applications such as VoIP. We think that the appropriate regulatory response is not to inhibit the ability of network operators to offer different levels of QoS. IP-based networks have been able to offer IP-based service at different levels of quality (typically expressed in terms of the mean and variance of delay. and also to ensure that NRAs make consistent use of the tools already present in the regulatory framework to address any market power problems that might emerge. and of monitoring compliance with the terms of those agreements. as real time bi-directional5 IP-based voice and video services have become increasingly commercially significant. however.6 These capabilities have been routinely deployed within IP-based networks. different levels of QoS generally enhance welfare. Some experts have expressed concerns over the use of different levels of QoS among networks (one of several possible concerns over network neutrality). these challenges will not necessarily go away. DiffServ and ATM were often used. however. simpler and cheaper solutions are often possible. .

Should intervention related to the price of IP data interconnection prove to be required. however. because (1) it is not clear how it would be justified under the current regulatory framework. and (2) a glide path would be necessary in any case in order to enable operators to adjust their retail payment plans over time to match the changing wholesale arrangements. NRAs must be able to intervene if interconnection breaks down. it minimises economic distortions.7 and also makes it possible to apply differential network-to-network interconnection prices to reflect different levels of Quality of Service. especially where this is a manifestation of some form of market power. it should address the aggregate level of prices. a principle known as Ramsey-Boiteux pricing. . it is desirable for the structure of retail prices to reflect the degree to which customer demand responds to price. This could be implemented by (1) accelerating the speed with which the maximum call termination rate declines from year to year under existing CPNP arrangements. o In an ideal world. or (2) by requiring all fixed and mobile operators to eliminate call termination fees altogether. We nonetheless stop short of recommending the second option. or (3) by permitting negotiated termination fees subject to an obligation that the fees be reciprocal (the same in both directions) between each pair of interconnected (fixed or mobile) networks. this power to intervene must include the ability to cap the price of IP data interconnection. This enables the operator to choose an efficient pricing structure. • o 7 In general. The main conclusions are: • We do not advocate an interconnection obligation as regards IP data traffic in general. and it involves the fewest impediments to the evolution over time of interconnection arrangements as networks evolve to an IP basis. we think that societal welfare would be substantially enhanced if mobile termination rates were much lower than they are today. The outright elimination of call termination fees is simple. but leave operators free to choose the structure of prices. so as to reasonably quickly achieve levels much lower than those that pertain today. and we do not see a need to mandate any-to-any peering. and possibly no higher than the rates that prevail today for fixed termination rates. the second option would appear to offer many advantages in comparison to the other two. To be effective. Independent of the migration to NGN.Final Report: The Future of IP Interconnection XIII Conclusions and recommendations Our conclusions and recommendations are presented more fully in the body of the report.

This fast glide path gives the operators time to adjust their business plans. We note that zero is in many respects an ideal number. one could reasonably expect that flat rate retail plans would emerge that included off-net calls to mobile phones. This is not to suggest that a new remedy is needed. it is possible that some operators would respond by attempting to capture economic rents from other operators in some other way. It is possible to degrade the quality an IP interconnection in subtle. attempts by operators to preserve high rates would tend to distort the migration. Reduction in mobile termination rates would tend to lead to a substantial decrease in usage-based retail prices for mobile service.10 It may be necessary for NRAs to take action from time to time (possibly ex post) to ensure that operators subject to an SMP remedy do not take actions that would render the IP-based voice service interconnection ineffective or unusable. at the expense of a significantly smaller increase in the fixed portion of the retail price9 of mobile service (the “waterbed effect”). difficult-to-detect ways. We are not indicating exactly what the target levels for fixed and mobile voice termination should be. or the quality of the IP interconnection that will have become the basis for voice interconnection. There is a strong argument that this lowering of mobile termination rates should be implemented before the migration to IP-based NGNs. 9 Initial fees and monthly fees if any would tend to be higher. to increased availability and consumer take-up of flat rate and “buckets of minutes” retail plans. since this is far beyond the scope of this study. it means that NRAs must be empowered to ensure that standard interconnection 8 Under the authority provided by Article 19 of the Framework Directive. We also observe that. . That seems to us to be a matter for the market to decide. or common with it. 10 This concern is largely independent of whether the interconnection to the IP voice service is physically distinct from the IP data interconnection. the terms and conditions. This corresponds to option 1. for reasons noted above. handset subsidies would be lower. It is conceivable that this might lead to manipulation of the prices. otherwise. rather. and allows for mid-course corrections if necessary. were mobile termination fees to be in the same range as current fixed termination fees. o o o • If call termination fees were in fact either eliminated or greatly reduced.XIV Final Report: The Future of IP Interconnection o What we concretely recommend instead is that the Commission mandate8 that fixed and mobile call termination rates “fast glide” to prespecified target levels over a predefined number of years (somewhere between three and five). and to substantially increased use (expressed in minutes of use per month) of mobile telephones.

If a network operator with SMP wishes to close Points of Interconnection. it should do so in consultation with competitors and with appropriate notice. NRAs and NCAs need to be prepared to address wilful deviations from network neutrality. absent a compelling need for retail remedies. (2) continued research and • • • 11 Numerous recent papers have taken up the question of the relative desirability of Calling Party Pays (CPP) retail arrangements versus Receiving Party Pays (RPP). we do not advocate major regulatory initiatives at this time. Existing SMP remedies should be kept in place for a reasonable transition period. technology and standards appear to be evolving satisfactorily. . We see merit in the use of Articles 20 and 22 of the Universal Service Directive to require ECSPs to document their practices as regards blocking access to services or degrading the quality of access to services. Four areas where policy initiatives may be appropriate are (1) considering techniques to address or ameliorate the consequences of the exhaustion of IPv4 addresses. especially where an element of economic foreclosure appears to be present. and thus well in advance of the full blown deployment of IPv6. the choice of retail payment model (for example. There may also be merit in enabling NRAs to mandate a minimum quality of service. CPP versus flat rate) should be a matter for the service provider to determine (reflecting the preferences of its customers). as the Commission has proposed. and is moreover altogether unnecessary in this report – this choice should be a market outcome. and no public policy response is required. but rather some variant of a flat rate plan. not something for the regulator to mandate. We think that this discussion is somewhat off the mark. It might be sufficient to require that the overall quality of IP-based voice interconnection that an SMP voice operator provides to competitors cannot be unjustifiably inferior to that which it supplies to itself (a condition more akin to equivalence of input than to non-discrimination).11 In regard to network neutrality. 12 Possibly to include the Regional Internet Registries (RIRs) enabling a secondary market for IPv4 addresses.12 which (after several false alarms) now seems to be quite likely in the 2010-2011 time frame. • Any necessary remedies should operate at the wholesale level. In most respects. until supplanted in most cases by new SMP remedies (to the extent that SMP is still applicable). The migration to an IP-based NGN inevitably implies a host of transition challenges. In particular. what most customers want is neither CPP nor RPP. In any case. the existing regulatory framework for electronic communications probably provides adequate tools. and competition law provides additional mechanisms.Final Report: The Future of IP Interconnection XV remedies (updated to deal with new technology) are not subverted. however. We generally consider the approach that a number of Member States have taken to be appropriate.

(3) research into the longer term issue of unbounded growth (“bloat”) of the BGP global routing table. and (4) research into technologies that could over time ameliorate the termination monopoly. and for measurement of compliance to those classes.XVI Final Report: The Future of IP Interconnection standards development in regard to classes of service for QoS. such as user ENUM (rather than carrier ENUM) or techniques to enable multiple operators to serve calls placed to a single telephone number. .

and the International Telecommunications Union (ITU)16. including those conducted on behalf of the German Federal Network Agency. one can gain a high-level overview of the report. A key element of this success has been the availability of a global IP-based network. See Elixmann et al (2007). and any errors. are our own. Marcus (2007). Interconnection on page 4. Some readers may wish to skim the concept boxes first. the Internet. See ERG (2007a). a number of technological. Electronic communications services based on the Internet Protocol (IP) have been incredibly successful. These network operators voluntarily agreed to connect their networks so as to form a global network – no significant regulatory intervention has been required to date in order to drive the interconnection of IP-based networks. comprised of many independent network operators. we have highlighted key concepts in boxed comments like this one. or Bundesnetzagentur13. IP interconnection has also been featured prominently in one report after another in the context of regulatory challenges as networks evolve to Next Generation Networks. the OECD15. and then delve into the full text of the report. By skipping from box to box. Berg et al (2006). market. Today. See Marcus (2006b). others may prefer to glance at the concept boxes as they work their way through the full document. the European Regulators Group (ERG)14. The concept boxes may be particularly useful to the uninitiated or novice reader. Marcus (2006a). Each key concept box contains a reference (a “hot link”) to the next one. . societal and regulatory factors are converging in such a way as to raise profound questions about how interconnection arrangements should work in the future. the Hungarian NRA. Key Concept Box 1: Note to the reader In this report.17 13 14 15 16 17 See Vogelsang (2006). Hackbarth/Kulenkampff (2006). for example in the recent report of the NHH. with the support of a number of distinguished outside experts. Continued in Access vs. The views expressed. These developments have motivated several important studies of interconnection in the evolving IP-based environment over the past two years. Experience in the Internet has been profoundly different in this regard from that in the Public Switched Telephone Network (PSTN/PLMN). See Paltridge (2006).Final Report: The Future of IP Interconnection 1 1 Introduction This is the Final Report of a consulting study on “The Future of IP Interconnection” which has been undertaken for the European Commission by WIK-Consult GmbH.

and with data transmission continue to decline. with data storage. Service providers can provide services. or video) can be provided over a single network. A set of recommendations as to how IP interconnection should be addressed in a regulatory context. the European Commission has requested: • An analysis and comparison of traditional interconnection models based on circuit switched technology.1 The changing electronic communications environment Notable among the technological. As a result. Finally. • • . it explains the structure of the balance of the report.2 Final Report: The Future of IP Interconnection For this report. In the case of data transmission. improvements based on Dense Wave Division Multiplexing (DWDM) could result in even faster improvements in price/performance. multiple applications (such as real-time voice. even if they are not network operators. For computation and data storage. The unit costs associated with computation. by way of noting that current arrangements leave much to be desired even in today’s PSTN/PLMN world. data is digital rather than analogue to a progressively increasing degree. explains why the future evolution of IP interconnection is a challenge. it is possible that substantially all electronic communications will be IP-based. • • This introductory chapter seeks to motivate the discussion that follows. but not the ability to offer it. It provides a brief overview of the economics of interconnection. It then touches briefly on the welfare implications of interconnection arrangements. It quickly reviews the objectives of the European regulatory framework and discusses their relevance to IP interconnection. 1. they have tended to improve by a factor of 4 every three years (Moore’s Law). and of IP interconnection models that are used in Europe and other world regions. As a related phenomenon. market and societal changes that motivate this study are: • All networks appear to be migrating from a circuit-switched technological basis to an IP packet-switched technological basis – in the future. including applications that were historically tied to a specific underlying network. The character of the underlying network may affect the efficiency and cost of the application. A forward-looking analysis of IP interconnection regimes in the light of the evolution of both the traditional telecommunications and Internet environments. It explains the methodology of the study. It notes the key drivers of change in the electronic communications environment.

As described below in section 1. Increasingly. but they are not the same thing.2 The economics of Interconnection today Network interconnection has evolved historically in the context of the “traditional” Public Switched Telephone Network (the PSTN/PLMN) and of the Public Land Mobile Network (the PLMN). Individual consumers can develop their own application content – consumers are not exclusively dependent on commercial content providers. 19 The European regulatory framework also leaves open the possibility that regulatory intervention to ensure interconnection might occasionally be necessary even where Significant Market Power is not present. different interconnection arrangements have prevailed. there are also a few examples of successful large scale voluntary arrangements in the PSTN/PLMN world.3. and Singapore.19 In the world of the Internet Protocol (IP). Voluntary negotiated arrangements have tended to form the core of this system. This section provides a brief overview of key issues and considerations. the latter is the mobile network.Final Report: The Future of IP Interconnection 3 • In this environment. 1. these new communications services are likely to link devices to people and to one another. notably in the United States. where only one party to the interconnection has Significant Market Power (SMP). 18 The former is the fixed network. electronic communications have served to link people to one another. • • 1. See Article 5 of the Access and Interconnection Directive. regulation is usually needed to ensure interconnection and to address asymmetries in bargaining power. Historically. not with access. The two are closely linked in popular perception and in the literature. new communicating applications can be brought to the mass market far more quickly than in the past. A comprehensive discussion appears in Chapter 3. however.2.2.1 The distinction between access and interconnection This report is concerned with interconnection. Canada. . and regulatory intervention has only rarely been attempted.18 It has been widely recognized that.

interconnection enables an operator to establish and maintain communications with the customers of another operator. to another undertaking. Continued in Call termination fees and the migration to IP on page 6. access to relevant software systems including operational support systems. p. Laffont/Tirole (2000). or to access services provided by another undertaking. access to fixed and mobile networks. Interconnection. e. for the purpose of providing electronic communications services. Interconnection enables an operator to establish and maintain communications with the customers of another operator. Interconnection is a specific type of access implemented between public network operators …“ 23 See Marcus (2007). ducts and masts. can be viewed as being a symmetric relationship between two operators. access to physical infrastructure including buildings. 20 Directive 2002/19/EC. by contrast.. The termination payment is intended to approximate the terminating network’s usage-based unit costs. Interconnection Access enables an operator to utilize the facilities of another operator in the furtherance of its own business and in the service of its own customers.4 Final Report: The Future of IP Interconnection Key Concept Box 2: Access vs. under defined conditions. access to virtual network services…” 22 The Directive continues: “Services may be provided by the parties involved or other parties who have access to the network. . Access is therefore often referred to as one-way access and “interconnection” as two-way access.”22 For our purposes. Access can thus be viewed as an inherently asymmetric relationship.”21 It defines interconnection as “… physical and logical linking of public communications networks used by the same or a different undertaking in order to allow the users of one undertaking to communicate with users of the same or another undertaking. access to conditional access systems for digital television services. in that one operator is providing access to another. which may involve the connection of equipment. 21 The Directive goes on to say: “It covers inter alia: access to network elements and associated facilities. on either an exclusive or non-exclusive basis.2 Current European arrangements for switched network interconnection Substantially all European PSTN/PLMN (fixed and mobile) interconnection follows a set of arrangements known as Calling Party’s Network Pays (CPNP).23 1. See. access to number translation or systems offering equivalent functionality.2. by fixed or non-fixed means (in particular this includes access to the local loop and to facilities and services necessary to provide services over the local loop). in particular for roaming. where the network of the party who placed the call (the originating network) makes a payment to the network of the party that received the call (the terminating network). The Access and Interconnection Directive20 defines access as “…the making available of facilities and/or services. while access enables an operator to utilize the facilities of another operator in the furtherance of its own business and in the service of its own customers.g. 5.

2.4. for example. and exceed by a still greater margin the level that would maximise overall welfare.25 as a result of this large differential. For this to be true. property rights of the parties involved need to be clearly defined. Coase was thinking of spectrum management. presumably distorting the evolution of both. a ratio that cannot be justified by differences in cost. the ideas are potentially just as relevant to interconnection.26 This results in a net transfer from fixed networks to mobile.. but not invariably. and even in otherwise competitive markets). Negotiated arrangements among Internet peer backbone networks have been commonplace from the first days of the commercial Internet in 1995. 25-32. Canada. 24 25 26 27 See Section 5. CPNP arrangements imply the need to maintain regulation indefinitely.Final Report: The Future of IP Interconnection 5 The implications of these payments are complex. calls from fixed phones to mobile transfer substantial sums of money to mobile operators.2. certain preconditions have to be met – for example. Regulated mobile termination prices in Europe have been on a glide path downward for several years now. while calls from mobile phones to fixed return only a small fraction of that money. however. These negotiated arrangements have frequently. 1.1.chapter 14. See Marcus (1999). and his work led to spectrum auctions and to liberalised spectrum management in general. thus. There are similarities between PSTN/PLMN negotiated interconnection arrangements and Internet negotiated interconnection arrangements. Ronald Coase (who subsequently received a Nobel prize in economics) argued that private negotiated arrangements are frequently superior to regulated arrangements. the United States.3 Negotiated arrangements In his landmark 1959 paper. but there are also significant differences. What accounts for the differences? Privately negotiated arrangements are frequently superior to regulated arrangements. and Singapore. a system sometimes referred to as Bill and Keep or Sender Keeps All. but nonetheless in our view still greatly exceed usage-based marginal cost. See Bomsel et al (2003). led to agreements to interconnect and to exchange traffic without payment.27 Negotiated arrangements have also long been the norm for PSTN/PLMN interconnection among mobile operators in many countries and also among non-dominant fixed operators in. See EU-Commission (2007).24 European termination fees on mobile networks tend to be about nine times greater than those on fixed networks at present. One of our major thrusts in this report has been to try to develop a deeper understanding of these distinct Coasian interconnection regimes. pp. It is generally accepted that operators would be motivated to set these termination payments at inappropriately high levels in the absence of regulation (even small operators. .

or that otherwise reduces the importance of the telephone number as a bottleneck resource. and even more so in an IP-based world where the direction of the call can easily be reversed. will continue to be problematic whether delivered by a conventional switched network or by an IP-based network.6 Final Report: The Future of IP Interconnection Given (1) that it is widely acknowledged that “traditional” CPNP arrangements will at best be severely stressed by the evolution to IP-based networks. the payment no longer represents the cost of the network. and data services have all converged so as to be delivered to European consumers over a common set of ultra high speed broadband access arrangements. Key Concept Box 3: Call termination fees and the migration to IP The migration to IP places stress on existing call termination arrangements: • The call termination fee is intended to reflect costs of the network. video services (television). unfortunately. It is entirely possible that regulatory and competition problems will emerge in regard to applications other than real-time voice. we see no serious prospect of a Coasian solution for voice services until and unless technology is widespread that enables more than one operator to deliver calls to a single phone number. and we see no need to take action against threats that for now are merely speculative. If the service provider is not the same company as the network provider (as can readily be the case in an IP-based NGN). Voice traffic represents a small and declining fraction of the cost of an IP-based NGN. As long as Europe uses CPP/CPNP retail and wholesale payment arrangements. . we have considered carefully whether Coasian arrangements might represent a more future-proof approach to IP interconnection going forward. we think that today’s Coasian arrangements will continue to be workable in general going forward. • • • Continued in Call termination fees: Mobile networks on page 14. To the extent that wholesale termination fees are unrelated to cost. The migration to IP will not in and of itself ameliorate the termination monopoly as long as only a single operator can complete a call to a given telephone number. (2) that these Coasian systems represent the primary alternative model that is known to work. but it is associated with a service. The voice service. The payment also assumes that the party that placed the call is the sole beneficiary of the call and should bear the total retail cost of the call. they will invite arbitrage that seeks to correct for the irrationalities in the pricing system. but they are not in evidence today. and we have explored changes that might be necessary to equip a Coasian system appropriately for IP interconnection in a world where telephony services. Minutes of voice use will have less and less relation to the cost of the network going forward. and (3) that they have already been shown to work well for IP-based traffic. For IP data services in general. The rationality of this assumption is dubious in the current network.

however. 4 April 2006.” One key foundation stone to their analysis is their claim that the migration to IP will solve the call termination problem.7. referred to as GRX in the current generation. provides IP interconnection and service interworking between mobile networks. . We have explored different regulatory regimes in the hope of finding less intrusive mechanisms.5G and 3G roaming and interworking. 29 See GSM Association. as well as MMS interworking.30 By implementing a common traffic interchange fabric among mobile operators. That claim is. as would be prescribed by a mandated charging model. A single wholesale model will constrain the variety of retail models that are necessary for efficiency. reflecting the volume of traffic each operator has carried and the Quality of Service at which it has been carried.29 The inter-PLMN backbone. rather than less. The IPX is intended to provide accounting and charging arrangements capable of “cascading biling”. it is not necessarily appropriate for regulators to mandate this model. We find that a regulatory response to the termination monopoly on the voice service will continue to be required for the foreseeable future. not at the level of IP packet – re-routing at the IP level has no more effect than re-routing at the circuit level. With the multitude of products being developed in the IP environment it would appear that operators will need more freedom to negotiate interconnection charges tailored to their situation. but we see no realistic prospect of totally withdrawing regulation of the voice service in the near or intermediate term. thanks to IP’s inherent capabilities to route traffic over multiple paths. The termination monopoly persists. As long as a call to a single telephone number must be served by a single operator.. the termination monopoly is likely to persist.4 The GSM Association’s IP-based Interconnection model A consulting study for the GSM Association has recently propounded a set of recommendations as regards IP Interconnection in an NGN environment.2. release 3. The bottleneck resource is at the level of the telephone number. Inter-PLMN Backbone Guidelines. 28 See Reynolds et al (2007). IPX is an upgrade to GRX that supports 2. coupled with some specific analysis of the conditions under which regulatory intervention might be required: “…[E]ven if a particular charging model develops commercial currency. This is a positive outcome. 30 See Inter-PLMN Backbone Guidelines and also Reynolds et al (2007).Final Report: The Future of IP Interconnection 7 1.28 They propose the use of Coasian arrangements. simply incorrect. the GSM Association has developed a series of standards for interconnection of mobile (PLMN) networks. even though the technology has been routinely deployed within individual Internet Service Provider (ISP) networks for perhaps a decade. it is possible that GSM-A will overcome the transition cost issues that have tended to inhibit the deployment of QoS among Internet providers. More concretely.

Applications and the associated Transport mechanisms 31 For general background on layered network protocols. see Tanenbaum (2003). As such. the endeavour raises a number of potential public policy concerns. or more recently WiFi). it can be viewed as an attempt to lock in today’s inefficient pricing structures. typically running in a single device – a server computer. and also of the Next Generation Network (NGN). video or data over cable. and (4) the changing cost structure of the network. Moreover. dial-up lines. which is a manifestation of convergence. which is the foundation of the Internet. telecommunications or wireless – has profound implications for the evolution of competition. We deal with this subject at length in Chapter 4.1. (2) the migration to Next Generation Networks (NGN).2 of this report.3 Emerging challenges to IP network interconnection A number of factors suggest that fixed and mobile switched interconnection and IP interconnection will all need substantial re-thinking in the years ahead. new applications and new transmission technologies can be introduced without disrupting existing arrangements. 1. and for regulation generally. Among them are: (1) the decoupling of the network from the service.31 In a layered network.. Each vertical “stack” represent software. leased lines. the IPX is a billing model rather than a technical model. 1. a single transmission link can carry any application. even as the network evolves in directions that would otherwise make them less relevant. To the extent that common interconnection arrangements facilitate coordination among mobile operators. they inevitably raise regulatory and competition policy concerns.4. (3) the evolution of the access network to an all-IP basis. To a significant degree. . or for that matter can simultaneously carry any mix of applications.8 Final Report: The Future of IP Interconnection At the same time.32 Figure 1 below shows the TCP/IP reference model.3. a router. In the early eighties. IP is the pre-eminent example today of a layered network protocol.1 Decoupling of the network from the service The decoupling of the network from the service – the ability to offer any combination of voice. IP is a key element of the “glue” that binds the applications to the underlying networks. We return to these points in section 5. and serve to motivate the current study. computer scientists developed the notion of layered data communications protocols in order to insulate data applications (such as the protocols associated with the worldwide web) from the transmission facilities over which they operate (such as Ethernet local area networks. 32 Whether the link can carry the traffic with the desired performance characteristics is a separate question. a personal computer (PC).

Final Report: The Future of IP Interconnection 9 provide useful services to end-users – for example. The IP layer. or some wireless technology). They can all operate over a single IPbased network connection (and it scarcely matters whether this connection takes place over the telecommunications network. This migration has profound implications for network evolution. e-mail. The Data Link and Physical layers are concerned with physical transmission of packets over wires. and thus for regulation. . and possibly a VoIP client. which is crucial. whatever. radio. web browsing. Figure 1: The TCP/IP Reference Model Server Router Personal Computer Application Transport Network Data Link Physical Network Data Link Physical Application Transport Network Data Link Physical Source: wik-Consult Routers are relatively simple devices that exist solely to forward information – they are not in principle involved with the application. You can invoke them whenever you want. and can run more than one of them at once. in your PC. You probably have an e-mail client in your PC. Note that data packets work their way up the stack to the IP layer. The first is that the application resides in large part in systems at the edge of the network – for example. forwards information from one system to the next. An IP-based network could contain dozens of hops. and then back down again in order to be reach the next “hop”. Voice over IP (VoIP). and a web browser. IP television (IPTV). Each Application talks to its corresponding “partner” Application in other computers. and could use different technologies to connect its routers. A router appears in the middle of Figure 1. cable television. but the applications implement this (horizontal) communication by requesting services from successively lower layers in the stack. which is why no Application or Transport layer is shown.

The network operator need not be the voice service provider – it might simply be forwarding packets. Will network operators welcome this change with open arms? Suppose your telecommunications incumbent provides your DSL broadband – will they be delighted if you make all of your telephone calls using Skype (an independent VoIP services that is either cheap or free. In this context. telephony and television. Analogously. For other services.10 Final Report: The Future of IP Interconnection Figure 2: IP-based applications running in your Personal Computer (PC) Personal Computer E-mail Client Web Browser VoIP Client IPTV Client Network Data Link + Physical Source: wik-Consult This fundamentally alters the job of the network operator.4. We return to this point in Section 4. or VoIP). the network operator is only involved in IP forwarding of packets. rather than their own voice services? Probably not! So they might be motivated to slow or to outright block this evolution if they can. for example. unidirectional video is IPTV. There is thus the risk that network operators who have market power might try to impede the ability of others to offer services that compete with their own – this is at the core of the Network Neutrality debate. but provides little or no revenue to the incumbent). it could instead be offered by some third party. This capability clearly exists in terms of the technology. but not at all in the P2P service. . In the case of peer-to-peer (P2P) services. depending on how you use it. The degree to which consumers will really enjoy increased choice under NGN is thus at issue. which once were closely linked to the networks over which they operated. Thus. there is likely to be a server somewhere – but while that server might be offered by the network operator. real time bidirectional voice is just an IP-based application (referred to as Voice over IP. can now be offered as network-independent services.

an NGN is IP-based. only a handful of points of interconnection (PoI) are strictly necessary from a technical and economic perspective. but their evolution is not very different from those of their counterparts in Europe. Still others. and in some cases their mobile networks. These different scenarios may have somewhat distinct implications for regulation in general. to Next Generation Networks (NGNs). is a question to which we return in Chapter 5. and if so for how long. notably BT. By contrast. Notably. possibly including DTAG. It manifests slightly differently for interconnection. Thus.3. such as KPN. might place greater emphasis on upgrading the access network to IP-based facilities that bring fibre close to the subscriber’s home. and for interconnection in particular. For an incumbent in a typical Member State. Whether they should be maintained. 33 See Elixmann et al (2007). if not into it outright. Many more are likely to in the years to come. Some operators.33 NGNs will not all evolve at the same time. while others will look to use NGN (often in conjunction with the Integrated Multimedia System [IMS]) to integrate their fixed and mobile services. apparently plan to upgrade the core of the network to NGN but not necessarily to make drastic changes to the access network. This potentially implies a complex transition for competitors. but they have tended to focus on this problem in the context of access rather than interconnection.S. IP-based interconnection facilities could potentially be much more concentrated than PSTN interconnection facilities.2 Migration to NGN A number of prominent European incumbents (including BT and KPN) have announced plans to evolve their existing PSTN fixed networks. An NGN delivers a wide range of services. including the full range of capabilities that we expect from fixed and mobile switched networks today. . it is sometimes associated with plans to phase out the existing PSTN altogether. perhaps as few as three or four. Some operators. Moreover. do not speak of NGN. there can be hundreds of PoIs for the PSTN. The physical facilities that support traditional fixed and mobile switched interconnection today would not necessarily exist in these NGNs unless regulators insist that they be maintained. Operators in the U. the migration to NGN potentially represents a major acceleration of the trend to base future electronic communications on IP. and possibly some stranded investment. are committed to upgrading the core and the access at once. A number of regulatory proceedings have considered the reduction in the number of Points of Interconnection (PoIs).Final Report: The Future of IP Interconnection 11 1. Some emphasize only the fixed network. or in the same way.

4 Changing cost structure All indications are that the emerging IP-based networks when fully deployed will have lower operating costs than existing circuit switched networks.3.34 This once again has regulatory and public policy implications. with or without NGN in the formal sense. WIK recommended instead that rates should.36 There is thus some debate as to whether a glide path or a step function is preferable in general. In the nearer term. 34 Some operators may. Implicitly. cost-based fees should follow this roller coaster ride. operators in the midst of transition will simultaneously bear the costs of both the old network and the new. their operating costs will not be as low as the eventual operating costs when transition is complete. it is possible that operating costs will increase before they decline. . that is a key driver to migrate an existing network to an IP base. it might be most appropriate to operate the subscriber line on an IP-only basis. 1. To the extent that interconnection fees are intended to reflect unit costs.3. particularly as regards the likely future evolution of competition.3 The evolution of the access network Today. at a suitable point in time. On a unit cost basis. there will no longer be a direct functional reason to maintain traditional voice capabilities on the subscriber line.5. 36 See WIK (2005). 35 See Ofcom (2005) and also Section 5.3. Ofcom has proposed to deal with this by ensuring that narrowband interconnection fees follow a glide path downwards. this implies challenges for the regulator. and in the absence of business drivers to the contrary we expect that it will be increasingly common to do so. Indeed. In technical and economic terms. simply be set to the forward-looking unit cost of the NGN.12 Final Report: The Future of IP Interconnection 1. and possibly a DSL broadband connection using the high frequency capacity of the same line. residential subscribers typically have a voice telephone line. however. In fact. have technical or economic reasons to continue to provide voice using the conventional low frequency capabilities of the DSL line. but ensuring that they remain somewhat above the “blended” unit cost until migration is complete.35 In a 2005 study for the Australian regulator. As the network evolves to a pure IP base.

Some of these are beneficial. Mobile operators in particular are motivated to subsidize the price of handsets and to provide other incentives to consumers in the hope of capturing lucrative termination fees. The prevailing view has been that high termination fees encourage strong competition for individual customers. for some of the newer Member States. in the mobile network. however. We would also note that very little analysis has been done of the corresponding negative impact that one might expect on the pace of adoption of fixed services. and from developed countries to developing countries. the termination fee arrangements have provided large subsidies that may have accelerated adoption of mobile services. and for mobile-to-mobile. it is often the case that termination rates charged by both interconnected network operators are similar. but others appear to be damaging to welfare. quite aside from the evolution that the network is undergoing.39 37 This is often the case. and that the number of calls originated and terminated between a pair of providers are roughly comparable. termination fees sometimes aid in providing universal service. with nominal penetration in excess of 100%38. especially Littlechild (2006) and Dewenter/Haucap (2005). these implicit subsidies may still produce benefits. because there is generally a single fixed incumbent that is obliged to provide universal service. 38. however.4 Implications of interconnection payments for consumer and supplier welfare The current payment regime has complex consequences.37 in those cases. termination fees generally result in very little money changing hands at wholesale level. and consequently its termination fees will legitimately be higher. Globally. This is felt to accelerate the adoption of mobile services. 38 See EU-Commission (2007). The EU as whole. but not always.Final Report: The Future of IP Interconnection 13 1. There is also a tendency at the international level for significantly more calls to be placed from developed to developing countries. its costs will tend to be higher than those of operators whose territories comprise easy-to-serve urban areas. . 39 A few recent papers have questioned these assumptions. Thus. there is good reason to reconsider interconnection arrangements. Where an operator’s territory is remote or otherwise expensive to serve. does not need to stimulate adoption. These tendencies imply a net transfer of funds from areas of high teledensity to those of low teledensity. p. For fixed-to-fixed calls. These factors play a minimal role in the fixed network in Europe.

and promoting innovation. Continued in Migration to Next Generation Networks: Regulatory Objectives on page 15. price and quality and for avoiding distortions of competition lead directly to regulatory provisions that address harmful market power in all of its forms. derive maximum benefit in terms of choice. … This is a useful and comprehensive set of goals. price. Taken together. and quality. These negative factors are quite substantial. 1. it provides little guidance to the policymaker to the extent that there is tension among the goals.2. . it is useful to review the objectives of regulation. We return to these points in Section 3. The most directly applicable of these are: Article 8(2): (a) ensuring that users.2. the Framework Directive itself establishes a series of regulatory objectives40.14 Final Report: The Future of IP Interconnection Key Concept Box 4: Call termination fees: Mobile networks The fees that mobile operators receive for terminating calls are currently too high. the high retail prices that high termination fees encourage appear to severely depress the use of mobile phones and the amount of fixed-to-mobile (F2M) calling.3. In the context of the European regulatory framework. the mandates for optimising user choice. as is often the case. High mobile termination fees appear to contribute to high retail prices for consumers. the goal of encouraging investment and innovation can be interpreted as an admonition to avoid needless regulation. these provisions necessitate complex judgments on the part of the regulatory community. In particular. many of which bear directly on IP interconnection. and to the exclusion of calls to mobile phones from flat rate plans. including disabled users. and is 40 Directive 2002/21/EC. one must consider the impact on retail price and on the use of these electronic communications services. Article 8. This complex balancing act infuses many aspects of regulation. however. (c) encouraging efficient investment in infrastructure. Against the backdrop of these neutral or positive factors. depressing use. Worse. (b) ensuring that there is no distortion or restriction of competition in the electronic communications sector.5 Regulatory objectives Before embarking on an exploration of regulatory policy. however. and thus reducing welfare.

and quality. Key Concept Box 5: Migration to Next Generation Networks: Regulatory Objectives Key regulatory objectives include: • • • Ensuring that users derive maximum benefit in terms of choice. price. Ensuring that there is no distortion or restriction of competition. and Encouraging efficient investment in infrastructure. in particular requiring transparency of tariffs and conditions for using publicly available electronic communications services. and promoting innovation.Final Report: The Future of IP Interconnection 15 relevant in judging the degree to which various aspects of IP interconnection should be regulated going forward. and to consumer rights. for example. … Article 8(4): … (c) contributing to ensuring a high level of protection of personal data and privacy. and end-to-end connectivity. 41 See. our judgment is that market power in regard to interconnection will still be an issue that regulators need to address. Other relevant objectives relate to the European single market. Haucap/Marcus (2005). Among the relevant goals are: Article 8(3): (a) removing remaining obstacles to the provision of electronic communications networks. . however. there is no discrimination in the treatment of undertakings providing electronic communications networks and services. Continued in Technological and market evolution on page 18 In the realm of electronic communications. (b) encouraging the establishment and development of trans-European networks and the interoperability of pan-European services. (c) ensuring that. threats to competition have arisen more often in connection with access and interconnection than in any other area of regulation. The migration of voice communication services from a circuit-switched basis to a packet-switched basis will change the character of market power. associated facilities and services and electronic communications services at European level.41 These questions are therefore very central to this report. (d) promoting the provision of clear information. in similar circumstances.

16

Final Report: The Future of IP Interconnection

(e) addressing the needs of specific social groups, in particular disabled users; and (f) ensuring that the integrity and security of public communications networks are maintained.” The concern about discrimination in Article 8(3)(c) is, as we shall see, relevant to the entire debate about Network Neutrality (see Section 4.4).

1.6

Methodology

We conducted the study using a mixture of desk research, stakeholder interviews, indepth analysis, and formulation of recommendations.

1.6.1 Desk research
An extensive literature exists on the economics of interconnection in the fixed and mobile switched world42, and a smaller but still substantial literature on interconnection in the world of the Internet. Despite this substantial investment in economic research, significant gaps exist in the literature, notably as regards (1) the relationship between interconnection in switched networks and IP interconnection43, (2) the relationship of both models to likely interconnection arrangements in the IP-based multi-service networks that are emerging today, and (3) the differences among fixed and mobile switched interconnection regimes in different countries. In the course of our research, we also came to feel that the literature had not adequately considered the termination monopoly in the context of a (PATS) service somewhat de-coupled from the underlying network.

1.6.2 Stakeholder interviews
In the course of the study, we formally interviewed many knowledgeable market players, regulators, and academic experts in order to gain insight and to understand their views. We also had numerous informal discussions with colleagues in government, industry and academia. We met informally with the European Regulators’ Group (ERG). Much of this interaction was by means of email or telephone, but some was face to face.

42 Laffont/Rey/Tirole (1998a), Laffont/Rey/Tirole (1998b), and Armstrong (1998) are generally regarded as the groundbreaking works.. For a comprehensive review of the economic literature on interconnection, see Annex 2. 43 See, however, Laffont et al (2003).

Final Report: The Future of IP Interconnection

17

1.6.3 Consolidation and analysis of results, formulation of recommendations
We analyzed the interview results and compared them with the results of our desk research. We attempted to bridge any gaps in theory and in knowledge in order to develop a more comprehensive understanding of the likely evolution of IP interconnection, and the implications of regulatory innovations. Finally, armed with this background and based on our own assessment of likely developments, we developed near term and long term recommendations to the Commission as to the appropriate regulatory direction going forward.

1.7

Structure of this report

Chapter 2 describes the technological, market and societal forces that motivate a reexamination of IP interconnection arrangements at this time, and concludes with a overview of likely IP-based traffic trends going forward. Chapter 3 provides a comprehensive overview of the economics of interconnection arrangements in the fixed and mobile switched environment in various regions of the world, and in the Internet, both at a theoretical and at an empirical level. Chapter 4 delves into the special topic of Quality of Service in an IP-based environment, with particular emphasis on (1) the distinction between QoS within a network versus QoS between networks, (2) the likely root causes of slow deployment to date of QoS between networks, and (3) the Network Neutrality debate. Chapter 5 draws on both the economic background in Chapter 3 and the material on QoS in Chapter 4 in order to address IP interconnection for VoIP services, and IP interconnection in the context of Next Generation Networks. Finally, Chapter 6 provides conclusions and Chapter 7 presents our recommendations.

18

Final Report: The Future of IP Interconnection

2

Underlying drivers of change

This chapter provides the broad context in which changes are taking place, distinguishing between shorter term phenomena and longer term (the latter being out to 2018). Key drivers are noted: separation of service from network, progressive improvements in price/performance, the migration of services to IP, the deployment of NGNs, changes in the access network, and so on. Broader effects, such as changes in Key Concept Box 6: Technological and market evolution

The marketplace has been continually upended by rapid technological change. Some of the most important changes under way include: • The decoupling of the service from the network: In contrast to telecommunications networks in the past, services and the connectivity needed to provide them in the future need not be provided by a single integrated firm. • Ongoing price/performance improvements due to Moore’s Law are enabling better, faster, and cheaper communications by fiber optic transmission (e.g. DWDM) and optical switching and cross-connection. Breakthroughs are also possible in the years to come, but their exact nature is difficult or impossible to predict. • The fundamental changes in the delivery of voice and video services evolving to IPbased networks. Video is migrating to IP much as voice is, but the implications are not necessarily the same. Video has enormous bandwidth demands, unlike voice; however, much video use is unidirectional. Not all video will migrate to IP; conventional cable television and satellite are very efficient media for delivering linear video to multiple subscribers. To the extent that video is indeed carried over IP, there are significant implications for network design. • Network operators throughout Europe are replacing traditional circuit-switched networks with IP-based Next Generation Networks (NGNs). • Communication protocols used for IP routing between independently managed networks (Autonomous Systems, or AS) will continue to evolve and improve, which will facilitate the ability of the Internet to grow and to scale. • IP address exhaustion and gradual deployment of IP version 6 (IPv6) is likely to become a significant concern over the next few years, with probable exhaustion of IPv4 addresses in the 2010-2011 time frame. The generally accepted solution, migration to IPv6, has seen only scant deployment to date. • Network back-haul in carrier core networks has historically been accomplished by dedicated circuits and Asynchronous Transfer Mode (ATM). In the future, these communications are expected to be implemented to an increasing degree by means of enhanced “carrier grade” versions of the Ethernet Local Area Network (LAN). Ethernet equipment is simpler and less expensive. • Network operators are deploying fiber progressively deeper in the local loop, i.e., closer to the customer’s premises. Continued in Symmetric vs. asymmetric relationships on page 34.

home networking and the increasing use of user-provided content (e.g. Web 2.0) are also addressed. Finally, we attempt to distinguish likely short-term, medium-term and long-term developments.

we discuss the same developments in a Next Generation Network (NGN) context.1. and the network service layer.1.1 De-coupling of the service from the network In section 1. Interrogation-.Final Report: The Future of IP Interconnection 19 2. Breakthroughs are also possible over the ten year time frame that we have been asked to consider. the media layer. and that change is continuing.1 Technological and market evolution The marketplace has been continually upended by rapid technological change. coupled with progressive refinements are likely. Figure 3: The NGN layered structure Network Service Layer Softswitch Control Layer Media Gateway Media Layer Access and Transport Layer Enterprise Customers Remote Office / Small Office / Home Office Residential Users Mobile Users Source: Uebele and Verhoyen (2001). The following figure shows that from a functional perspective a NGN basically consists of four different layers: the access and transport layer. A key trend is the separation of the service from the network. in this section. but these are in their nature difficult or impossible to predict.Call State Control . Serving. The key element of an NGN (apart from the transport function) are softswitch functions.3. we discussed these developments from the perspective of the Internet Protocol (IP) in general. provided either by a central unit (call server or media gateway controller) or distributed over various functional elements (Proxy-. 2. the control layer. Ongoing improvements in price-performance.

Vonage. In particular. These new network-independent services.1 Third party service providers Historically. • It is still an open question whether network operators who have market power might try to impede the ability of others to offer services that compete with their own – this is at the core of the Network Neutrality debate. Google. on the other hand do not have (that much) an incentive to migrate their business to mere transport and routing capacities and functionalities. have profound implications.1. From an institutional perspective more or less integrated business models are likely to evolve in the next decade: • Traditional telco players (incumbents and competitors alike) of course have an incentive to keep the end user not only in the physical access and transport business but also with respect to services and applications. today. Media Gateway Control and Service Control.4. the competitive landscape is transformed. e-bay. Rather. Platform providers like e. independent VoIP service providers such as SIPgate. . The control functions can be specified as follows: • • • Call Control.e. 2. they are purchasing it from third parties as a wholesale service and they rely on the existing physical access and transport infrastructure that the end user is paying for with his/her monthly rental for the broadband access line. to avoid to be reduced to the mere pipe provider. IPTV (television over IP). inasmuch as competition to provide the consumer’s broadband access might be distinct from competition to provide the customer’s voice service over some broadband connection. which is still in its early days. and the changes in the entire value chain that they imply. only the telecoms network provider was in a position to offer telephony. which are responsible for signalling and the control of resources in the network.g. Amazon. The degree to which consumers will really enjoy increased choice under future IP-based networks is thus at issue.1. i.20 Final Report: The Future of IP Interconnection Functions) in the IMS frame. We return to this point in Section 4. and Skype offer services that substitute in greater or lesser degree for conventional telephony.com etc. has the analogous potential to provide an effective substitute to conventional television.

Telecoms providers are capable of providing voice. The differences in price and in performance may be greater in these cases. the same is true of cable operators. and to mobile service as well.Final Report: The Future of IP Interconnection 21 2. but the underlying principle is the same: In an all-IP world.1.1. Figure 4: Convergence of content delivery Narrowband World 1920s – 1990s Broadcast/Radio (RF Spectrum) Broadband World 1990s – present Broadcast/Radio (RF Spectrum) CATV (coaxial cable) Telecommunications (twisted copper wire) CATV (coaxial cable) Telecommunications (twisted copper wire) Voice Broadcast Video Broadcast Voice Broadcast Video Broadcast Broadband Data Choice Endusers Endusers Source: wik-Consult Telephony ISDN Broadband Data Multimedia Multimedia Multimedia Telephony Data Telephony Data Telephony . but fundamentally they are interchangeable. The same considerations apply in principle to fixed wireless access.2 Interchangeability of networks The de-coupling of the service from the network implies that the differences between networks based on distinct technologies matter much less than they used to. all transmission media can be fundamentally treated as “dumb pipes”. video and data. Platforms may have price-performance or other advantages for delivering one service or another.

1. The attractiveness to the customer is driven by convenience. many operators are motivated to offer as comprehensive a package of services as possible to the end-user. by offering for example triple play (voice. thus maintaining a virtuous cycle. often reinforced by competitive pricing that reflects the triple play provider’s economies of scope.720 1. The computers that use the network (and thus generate load) have gotten faster at about the same rate as the routers that provide the underlying network.800 1.1. all over a common connectivity platform) to a customer. The hope is that.2 Continued price/performance improvements due to Moore’s Law and to enhancements in fibre optic transmission (e.000 Intel Dual Core Itanium® processor 1. the network operator mixes higher mark-up services with low and at the same time reduces the risk that the customer will “churn” to another network operator.22 Final Report: The Future of IP Interconnection 2.g.200 Transistors 1.3 Tripe play / multiple play services In response to these trends. DWDM) and optical switching and cross-connection Figure 5: Introduction of Intel Chipsets 1992 -2006 Millions 2. 2. video and data.000 800 Intel® Xeon™ Processor MP 675 Intel® Itanium® 2 592 Intel® Pentium® 2 410 Pentium® 3 1993 1994 1995 1996 Pentium II Pentium III 24 8 1997 1998 1999 Pentium 4 Intel® Itanium® 2 42 220 Intel® Itanium® 55 2000 2001 2002 2003 2004 2005 2006 600 400 200 0 1992 Source: Intel Technological progress has lead to substantial improvements in the price/performance of computers and of communications devices over the past two decades.600 1.1. .400 1.

The Free Encyclopedia. Wave(length) Division Multiplexing (WDM): WDM is a technology allowing to multiplex multiple optical carrier signals on a single optical fibre by using different wavelengths (colours) of laser light to carry different signals. This trend enhances both price-performance and reliability. optical switching will be used only where traffic density is high enough to warrant. overall the prices “per unit” have decreased considerably. . At the same time. which is still amazing. 1. As much as possible. Gilder’s Law asserts that "bandwidth grows at least three times faster than computer power. and as with any exponential relationship in real life it is not going to hold forever. then communications power doubles every six months. see the next paragraph). Thus.6 Tbps46 over a single fibre pair. 46 A Gbps is 109 bits per second. 47 Wikipedia contributors. http://en. however. if computer power doubles every 18 months (Moore’s Law).6 Tbps is 160 times greater than 10 Gbps. Thus.47 Optical (lambda) switching: Wherever possible. the network core is designed today to switch at a purely optical level.Final Report: The Future of IP Interconnection 23 • Moore’s Law asserts that the number of transistors on an integrated circuit for minimum component cost doubles every 18 months. WDM systems allow carriers to expand network capacity without laying more fibre. and can thus expand a basic 10 Gbps fibre system to a theoretical total capacity of over 1. some predictions along these lines have been overblown – notably. Rather. Current WDM systems can handle up to 160 signals.php?title=Wavelength-division_multiplexing&oldid=134919128 (accessed June 8.” In other words. This effectively multiplies the capacity of each optical fibre by the number of wavelengths carried.wikipedia. however.44 Moore's Law is not a law of nature. which has meant that network capabilities have kept pace with bandwidth demand. Semiconductor product development cycles subsequently accelerated to the point where it was more nearly three years. 45 Around 2000. it conforms well with empirical evidence since it was formulated in 1965. For the foreseeable future. a Tbps is 1012 bits per second. fibre optic capacity has grown exponentially in parallel with computing capacity (largely as a result of DWDM. some estimates were that the number of wavelengths being carried on an optical fibre using DWDM was doubling every ten months. 2007). even if absolute prices have increased. "Wavelength-division multiplexing. • • • This significant increase in performance has not been accompanied by linear price increases. the network designer seeks to avoid unnecessary optical-electrical-optical (O-E-O) conversions. The assertion was probably never true45. Some examples might underline this: 44 The initial claim was that the number was quadrupling every four years." Wikipedia. of course.org/w/index. Gilder’s advocacy of a pure fibre “home run” between any two parties who wish to communicate.

3 Fundamental changes in the delivery of voice and video services The movement of voice service from the circuit-switched world represents a profound transformation. 2. Today.3. This is effectively $30 per GFLOP.000th of the price of nine years ago. .1. Several business models can be observed. The effective increase is less. due to increased operating system overhead and codebloat. Jorgenson and Vu find that the “… surge of IST investment [in the G7 from 1995-2000] resulted from a sharp acceleration in the rate of decline of prices of IST equipment and software. as semiconductor product cycles shrank from three years to two. • 2. or one 1.000 US.48 At a macroeconomic level. which can be explained on the basis of Figure 6. of course.24 Final Report: The Future of IP Interconnection • In 1997 a processing platform capable of providing 1 billion (109) Floating point Operations per second (GFLOP) cost about $30. Figure 6: The value chain for the provision of VoIP ITSP-Services Local Line Network Access Network IP Transport PSTN Ga w te ay BB-PoP End Customer Phone Number Terminal Equipment/ Softphone DSLAM IP-Platform Ga te w ay Other IPPlatform Source: wik-Consult 48 Ibid. an off-theshelf PC provides about 10 GFLOPs for about 300 US $.”49 They conclude that these rapidly declining prices reflect in turn an acceleration of Moore’s Law. 49 Jorgensen/ Vu (2007).1 Migration of voice to VoIP and more generally of multimedia to IP The migration of traditional TDM based voice traffic to IP based technology is already under way in many countries of the world. The impending migration of video may represent yet another.1.

prerequisite: broadband access and Internet access of any provider. separate provision of VoIP services. it is not clear that all video will migrate to IP. prerequisite: Broadband access and Internet access of any provider. but the implications are not necessarily the same. Video has enormous bandwidth demands. In the German market. provided under different QoS parameters ranging from a PSTN/ISDN quality standard (G. we find examples of the following alternative business models: • Fully-fledged infrastructure based provider: VoIP services bundled with broadband access based on fully owned network access facilities and Internet access. In all of these cases. “Skype”. pure VoIP offering. ISP. First. ISP. . VoIP services bundled with broadband access and Internet access. ISP. broadband access based on a resale product of the incumbent. Multiplexing in the frequency domain. much of the use is unidirectional. VoIP services bundled with Internet access. may continue to be the preferred solution for some video for some time to come. and using inherent multicast capabilities of the underlying medium. main difference: the provider is not an ISP.Final Report: The Future of IP Interconnection 25 With this figure. objective: full replacement of traditional PSTN incumbent access. customer is contractually obliged to keep his PSTN access line with the incumbent.1. and PON-based FTTH systems can offer similar capabilities to subscribers. unlike voice. VoIP offering also for other customers. • • • • • 2. however. as is the case with television today. one can identify a multitude of more-or-less facilities based VoIP business models.3. there is significant prospect that multiple subscribers will want to view the same video at the same time. Conventional cable television and satellite are very efficient media for delivering linear video to multiple subscribers. Moreover. the physical transmission path has already been optimized to a significant degree for video. services are identical with the services in the preceding category.711) up to services without any definite QoS.2 Evolution of video to an IP-based service Video is migrating to IP much as voice is. partly bundling with own broadband access facilities and Internet access. prerequisite: DSL access line from the incumbent. These characteristics have significant economic and practical implications. however. separate provision of VoIP services. Service Provider.

Insofar as possible. and that they work internally with virtual containers provided by the so called Generic Framing Procedure (GFP). In a recent report submitted to the German Ministry of Economics and Technology.3 Deployment of Next Generation Networks Many communications carriers – incumbents and competitors alike – all over the world are currently investing in NGN technology.3. 2. carriers can follow a piecemeal approach to migrate to a NGN. the deployment strategy could be characterized as an attempt to achieve the “best of both worlds”. video would tend to be cached today within the same network as the user – file transfer of video from the content provider to the cache would not require better-than-best-efforts QoS. the authors point out that there are basically two different approaches to migrate from the current network world to a unique next generation multi-service network51: • On the one hand. These systems often are based on NG-SDH (Next Generation Synchronous Digital Hierarchy). to the extent that video is indeed carried over IP. There are also questions as to the degree to which differentiated Quality of Service (in terms of delay and packet loss) will prove to be relevant to IP video that flows across the interconnect point. it is quite possible that the preferred service model will be to use QoS only within individual IP-based networks (if at all) rather than between networks. there are significant implications for network design. At present.1. If transmission capacity were to decline in unit price even faster than storage and processing power.50 the trade-off would tend to shift in favour of shipping video somewhat longer distances. as some have predicted. Given the huge data volumes associated with video. the operator can rely on already existing and paid-for network technology. unique virtual containers are reserved either for voice services or for packet oriented data services. which is to say that the network elements are linked over SDH technology. and thus to avoid sending video one stream at a time over large distances. conventional PSTN technologies are utilized in parallel with modern IP components. 50 See the discussion of Gilder’s Law in this chapter. . To this end. Also. 51 See Pohler/Beckert/Schefczyk (2007). In other words.52 Within the SDHframings. A key question is whether storage capacity is likely to improve in price-performance faster than transmission capacity.26 Final Report: The Future of IP Interconnection Second. streamed video does not require better-than-best-efforts QoS as long as the user is wiling to tolerate a second or two of delay when a video stream is first initiated. In this case. hybrid network elements are implemented containing upgrade options for broadband data technologies. 52 For an introduction into the concept of NG-SDH see Kartalopoulos (2004). Thus. a service provider’s preference would generally be to cache non-real-time video close to the viewers.

but no specific proposal has yet emerged that seems likely to attract enough support.1. as will the system of AS number identifiers used for purposes of exterior routing. Typically. and finally the related issues driven by Multi-Protocol Label Switching (MPLS). . and as conventional GSM-based mobile service otherwise. 2. Indeed.g. A number of researchers have been working on improved routing protocols for quite a while. particularly in the standards organizations. and utilized only for some exotic or legacy services. the first step of the migration towards an end-to-end IP based network is the transition of the backbone (core) network. faster. a new breed of services is evolving combining fixed-link services with mobile services. if new living areas are constructed. 2.Final Report: The Future of IP Interconnection 27 • The second migration variant rests on the establishment of an end-to-end IP infrastructure. or AS) in the Internet is accomplished by exterior routing. e. In the following sub-sections of the report. however. Border Gateway Protocol v4 (BGP4). then the AS numbers. will not be able to scale enough to meet the needs of the Internet a decade from now. Conventional wisdom holds that this solution is better. the network technology in the access networks often rests heavily on traditional PSTN technology. and it is based on real next generation broadband systems. the traditional narrowband world plays only a minor role. and more cost-efficient than the piecemeal migration.4. to become BGP4's successor – thus.1. In the final stage of migration to such an NGN access network.4 Gradual modernization of exterior IP routing Routing between independently managed networks (Autonomous Systems. This variant is used in particular in greenfield situations. Moreover. it is likely that BGP4 will remain the dominant inter-domain routing protocol for a long time to come. for example by offering handsets and respective services for the end user which function as WiFi-based VoIP where Fixed Wireless Access technology is available. Nonetheless. several carriers in the world are making this transition. The communication protocols used for this routing will continue to evolve. The system technology underlying this variant is called Multi-Service Access Node (MSAN) or Access Hub (AXH). we consider first the routing protocols. in many countries of the world carriers are in the process of deploying fibre optic technology “nearer” to the end user.1 Routing protocol enhancements (BGP) Many Internet researchers believe that the current inter-domain routing protocol.

c.2 Autonomous Systems Numbers BGP4.which may not be the path with the most capacity. claffy.54 It was originally designed to reduce the complexity and thus to improve the performance of routers in ISP backbones. . When Classless Inter-Domain Routing (CIDR) was introduced in the mid-nineties. an ISP can specify which path a particular stream of data traffic will take. and also to support traffic engineering. ASNs were originally defined as 16bit values. 2. and vendors are starting to implement them.1. At the current rate of ASN usage. CAIDA. Over time.3 Multi-Protocol Label Switching (MPLS) within and among networks (and the evolution to GMPLS and beyond) Multi Protocol Label Switching (MPLS) is a technology developed by the Internet Engineering Task Force (IETF). however. 2. “bushier”. Note that the migration to IP version 6 (IPv6) (see Section 2. use Autonomous System Numbers (ASNs) as identifiers for multi-homed sites or ISPs.28 Final Report: The Future of IP Interconnection The steady growth in complexity of the global Internet routing table poses a severe challenge to BGP routing. As one might expect. it exacerbates it. the revised software should be complete and deployed long before the current 16-bit ASNs run out. k. solutions that can be simply layered on topic of existing mechanisms have decent prospects of deployment. Traffic engineering is used to distribute traffic in a network in ways that normal IP routing will not. and more than half of the possible ASNs have already been assigned in the Internet. it initially resulted in a substantial simplification of the routing tables by organising them hierarchically.4. the Internet routing environment is becoming progressively more complex. This is causing a steady growth in the size and complexity of the global routing table. With traffic engineering. and less hierarchical.1. if they deploy at all. and thus be better able to match capacity to demand. The IETF has been working on 32-bit ASNs for a number of years ( which will increase the number of ASNs from about 64 thousand to about 4 billion). The standards are now mostly done. and some intra-domain routing protocols.6) does not solve the problem of routing table bloat – if anything. 54 It is based in large measure on earlier work by Cisco Systems. The problem is well understood.53 There have been a number of attempts over the years to enhance the security of the BGP routing protocols. Normal IP routing will send traffic along the shortest path to a destination -.1. but it has not been solved. 53 Private communication. while solutions that involve significant complexity or incompatibility are often slow to deploy.4.

One must be careful with this kind of hyperbole – the IPv6 address space is 16 used with incredibly low efficiency. 2. There are some significant complexities with interprovider traffic engineering. however. MPLS has been progressively extended to address traffic engineering. provisioning the traffic path.1. IPv6 can address about 2 times as many devices as IPv4. rather than IP routing. greater than the number of electrons in the universe. Still more recently.56 The specifications for IPv6 were for the most part developed from 1991 to 1994. 56 It is sometimes claimed that the IPv6 address range is. and have been progressively refined ever since.Final Report: The Future of IP Interconnection 29 High-end routers are now fast enough that the performance improvement provided by MPLS has turned out not to be an important factor. setting up cross-connects. The core specification for IPv6 has been stable since 1998. for example. not the least of which is the need to know more about your competitor's network than your competitor is comfortable with you knowing.55 There has been interest for many years in higher level routing functions. not all of which are available for general use. and especially in policy-based routing. In contrast. and allocating bandwidth from the paths for the user requested service.1. GMPLS allocates an optical path through the network. MPLS has been adapted to enable the support of optical switching. The actual path that the traffic will take through the network is not specified by the user. That is more than enough. and it is likely to become more so over time. not between or among ISPs. give or take a few powers of two. MPLS is now widely used for virtual private networks (VPNs) within a single ISP. IPv6 could potentially address up to 2128 devices. . 2. Service providers will likely be even more interested in such capabilities as the entire switched telephone network migrates to an IP base. MPLS-based VPN services generally operate only within a single ISP. A key component of this transformation has been the incorporation of traffic engineering into interior routing protocols (OSPF-TE and ISIS-TE). The ITU-T and the IETF are both working on inter-ISP MPLS interconnection technology and methodology. traffic engineering can be important. Even so. Today.57 55 Major research initiatives are in progress on the part of the European Commission and the U. in many of today's networks.S.6 Gradual deployment of IP version 6 (IPv6) The maximum number of devices addressable by IPv4 is 232 (about 4 billion). 57 See Deering/Hinden (1998).5 Future evolution of higher-level routing functions A number of research institutions are studying the future evolution of the Internet. National Science Foundation (NSF).

Faster processing of protocol header information. some immediate and some potential. it might be difficult or impossible to expand existing IP-based networks or to deploy new IP-based networks or services.30 Final Report: The Future of IP Interconnection The vast expansion of address space in IPv6 enables a number of advantages. . 58 IPv4 systems commonly also support IPSEC. To date. NGN IMS (ETSI TISPAN) supports both versions of IP: IPv4 and IPv6. consequently. The other benefits of IPv6 were not sufficient to drive widespread deployment. the prime motivation for IPv6 deployment failed to materialize. 3GPP IMS was designed to operate only under IP version 6. Simplified management of mobility. additional to simplify significant offers the Integrated support for the IPSEC security protocol (providing privacy and authentication). There have been a number of “false alarms” where address depletion was incorrectly predicted. IPv6 will be widely deployed before IPv4 addresses are depleted. In addition. IPv6 attempts and rationalise the IPv4 protocol design in a number of small but possibly ways. For example. IPv6 systems must support it. IPv6 potentially following advantages: • • • Simplified auto-configuration. In addition to the ability to address more devices. 59 See http://www.59 Some experts worry that hoarding of scarce IPv4 addresses could cause problems somewhat sooner.potaroo. A quite credible analysis by Geoff Huston shows the IPv4 address space exhausting in the 2010-2011 time frame. IPv4 systems may support IPSEC.58 An expanded number of code points for differentiated Quality of Service. If this is not the case. Lack of enthusiasm for IPv6 has been one of several reasons why most deployed IMS at present is non-standards-compliant IPv4-based “pre-IMS”.net/tools/ipv4/ . all indications are that address depletion really will be a significant factor over the next few years. systems that have sought to use IPv6 addresses have had mixed results. • • IPv6 has seen scant deployment to date. for example in 2005. Hopefully. By contrast. At this point. The depletion of IPv4 addresses that had been feared since the early nineties was averted through a number of address conservation measures.

and education) are already motivated to adopt IPv6 for reasons that are in many instances linked to their respective specific circumstances. the Korean government has placed great emphasis on IPv6 deployment as part of its “Ubiquitous Korea” national ICT strategy. Eurocontrol plans to use IPv6 for next generation Air Traffic Control. there is no native support for IPv4). Ethernet VLANs. 2. Point to point Ethernet Transport. In the future. Hierarchical Addressing. Indeed.e. defence. Moreover. First and foremost. and the ability to provide QoS (dealt with in the IEEE 802. Also. NATO is moving to IPv6 overall. Ethernet VLAN Identifier Switching.Final Report: The Future of IP Interconnection 31 A number of significant firms and organisations in specific vertical industries and sectors (including transportation. Many factors are driving this evolution.7 Changes in network back-haul In carrier networks. IPv6 is thus an integral component of the Korean government’s master plan for industry. Ethernet-based solutions can also offer a number of advantages. Optional add-on standards to Ethernet already address a number of key needs that were historically lacking. Ethernet QoS. and therefore inexpensive. Multi-level Administrative Domains. typically used in conjunction with 802. requiring agencies to take up IPv6 whenever they replace equipment and generally no later than 2008. and support for virtual private LANs (VLAN). these basic Level 2 communications are expected to be implemented to an increasing degree by means of enhanced “carrier grade” versions of the Ethernet Local Area Network (LAN).1p standard. Standards bodies are currently taking up eight additional Ethernet improvements that are expected to collectively ensure that Ethernet will be a central element in the future access architecture of NGN: • • • • • • • • Duplex Ethernet. Protection Switching. much of the communication at Level 2 of the OSI Reference Model (the Data Link Layer) has historically been accomplished by equipment that implements the Asynchronous Transfer Mode (ATM).1q or with bridging). including the ability to deal with much more widely distributed networks. . the Japanese government has actively promoted IPv6. including better multicast support. The Chinese CERNET higher education network is pure IPv6 (i. Ethernet equipment is simple.1.

Generally speaking. These characteristics are at least threefold: • • • Number of Main Distribution Frames (MDFs) Number of street cabinets Distribution of sub-loop lengths. there are also some characteristics of the current (narrow band TDM) network which have a severe impact on the decision which kind of fibre architecture a carrier should deploy. however. In Europe. the average subloop length is more than 700 metres. a VDSL strategy might not turn out to be useful because VDSL 2+ loses all of its advantages if the loop lengths are “too long”.e. i. In France. Several alternatives can be identified: • • FTTN: Fibre-To-The-Node. fiber is deployed up to the Street Cabinet FTTP: Fibre-To-The-Premises FTTB: Fibre-To-The-Building (also called Fibre-To-The-Basement or Fibre-ToThe-MDU (Multi-Dwelling-Unit) FTTH: Fibre-To-The-Home • • • The incentives for carriers to deploy one of these alternatives rest on many factors such as competition intensity and capital market conditions. while in Germany the ratio is 40:1 and in the Netherlands it is 20:1. however. which enables the incumbent to exploit the full bandwidth potential of VDSL to a substantial share of its customers. which is to say that they are inversely correlated. . these ratios differ a lot between the Member States: in France. for example. a low ratio of street cabinets/MDF implies (ceteris paribus) relatively low unit costs to deploy fibre to the street cabinet and to follow a VDSL (2+) strategy. which means that the bandwidth that can actually be delivered to end-users is not that much higher than with ADSL 2+. Roughly speaking. A relatively high ratio of street cabinets to MDFs tends to imply that loop lengths are short. the average sub-loop length is 300 metres. In Germany. fibre is deployed up to the MDF FTTC: Fibre-To-The-Cabinet or Fibre-To-The-Curb.32 Final Report: The Future of IP Interconnection 2. If the loop lengths are too great.1.8 Replacement of the circuit-switched access network with a high speed IPbased network Several carriers in the world are currently are already underway in migrating to NGN access technologies (or at least they have announced a migration). the ratio is 10:1. this means that fibre is deployed deeply in the local loop.

which historically charged for some of their highest value content.2. and then reflects the evolution of the home network. and a single server can support a great many clients.0.0 and peer-to-peer).com offers books for sale over the Internet. the ultimate goal is the same: to eventually migrate the network to IP over Ethernet over fibre. the technical implementation can be viewed as being client-server. Amazon. and this continues to be an important business model. newspapers such as the New York Times. often for free. and the shift to user-provided content The migration to IP-based services is substantially changing the value chain of services and in turn the character of business relationships at the application and services layer is changing over time. This section of the report considers first the shift to user-provided content (including Web 2. . now offers all of their content for free. there are changes in the way that this technology is being implemented in the home. many radio stations make their broadcasts available to the public. Peer-to-Peer (P2P). Historically. while in other countries they have an incentive to deploy FTTB/H technologies. most content and most application services were offered by commercial service providers. The relationship is asymmetric – the client’s job is not the same as that of the server.1 Web 2. 2. our hypothesis is that incumbents in some countries have an inherent incentive to start with FTTC/VDSL technologies. 2.2 Broader changes in the use of electronic communications In parallel with these broader trends in the service provider world.Final Report: The Future of IP Interconnection 33 To make a long story short. in the sense that the software running on the customer’s Personal Computer (PC) (often just a web browser) is the client of software running on a server platform of the service provider. In some of these cases. and also a profound transformation in the way in which consumers utilize electronic communication services. the customer “pays” by being subjected to advertising. In all of these cases. Whether the network operator chooses FTTC/VDSL or FTTB/H fibre infrastructure.

but basically the migration to NGN does not alter these dynamics. the server. Peer-to-Peer (P2P). . The content is shared among users. or by recording the event (for instance. 60 The TCP/IP connection model of the Internet inherently supports both models of application interconnection – both client-server and peer-to-peer. which has generally been viewed as appropriate under the fair use doctrine of copyright law? These are important public policy issues. but in the context of the present study they can be ignored. the users typically have a symmetric relationship with one another. these services are often implemented as peer to peer (P2P) systems. These capabilities raise profound public policy questions. accesses services on another networked computer. A single server can support a great many clients.0. by making an unauthorized recording of a live performance. and the shift to user-provided content on page 35. (2006) define Web 2. and share it with others over the Internet. The computers typically have a symmetric relationship with one another. This is an inherently asymmetric relationship – the functions of the two computers are not the same. Peer to peer (P2P) – a configuration where two or more networked computers jointly deliver a service. In these systems. At a technical level. or by pointing a video recorder at a screen display of a movie). Do these recordings suppress the incentives for artists to innovate? Or are they instead more akin to an individual’s use of a videocassette recorder (VCR) to record a television program. YouTube is a conspicuous example of this new development. 60 Where users have the ability to legally share their own materials.34 Final Report: The Future of IP Interconnection In newer business models. Continued in Web 2.0 as a philosophy of mutually maximizing collective intelligence and added value for each Key Concept Box 7: Symmetric vs. participant by formalized and dynamic information sharing and creation. the “customers” or users may provide their own content. asymmetric relationships Client-server – a configuration where one networked computer (often the user’s Personal Computer (PC). which is often referred to as Web 2. rather than emanating from a central server. Högg et al. the client. NGN/IMS could enhance these interactions (for example by communicating bandwidth requirements back to the underlying transmission platform).0. they also have all the technical capabilities necessary to share materials copyrighted by third parties – either by copying the recorded medium directly. a technical realization that better fits the traffic flows and control relationships that characterize these systems. perhaps best exemplified by the Napster case in the United States.

In these systems.de/web2. an individual user can determine via feeds when he downloads Key Concept Box 8: Web 2. wikis. Their WebCenter Suite does not only encompass wikis. and share it with others over the Internet. In newer business models.2. Moreover.0 components. Oracle is pursuing a somewhat different approach. and social bookmarking. Indeed. Web 2. the users (and their respective computers) typically have a symmetric peer-to-peer relationship with one another. Presumably. rather than emanating from a central server. The TCP/IP connection model of the Internet inherently supports both models of application interconnection – both client-server and peer-to-peer.. with Lotus Connections IBM offers functions such as blogs.0 62 See Schäfer (2005). enabling the use of RSS feeds.0 technologies will become core applications within many companies. . several software vendors are in the process of enhancing their core applications with Web 2. thus. knowledge management and internal cooperation (collaborative working). see www. the “customers” or users provide their own content.0 approaches might also have a market potential regarding companies. as are Lotus Notes and Microsoft Office. important information can be made available centrally via weblogs. Microsoft is introducing the SharePoint Server. Historically. often referred to as Web 2.2 The changing environment within the home Convergence of technologies and services will have a significant impact on people’s everyday lives in the coming years. social bookmarks and the like. This option might be reasonable if. Continued in Internet Traffic Growth on page 41. Indeed. one can basically distinguish the following main application areas:62 61 Indeed. Moreover. via “Connections for Partners“. Home networks (E-home) will play an important role in this development. For more information. and the shift to user-provided content The migration to IP-based services is substantially changing the value chain of services and in turn the character of business relationships at the application and services layer. tagging.0. blogs or discussion platforms.berlecon. Peer-to-Peer (P2P). a new product is developed jointly with external partners or suppliers. organisations external to a company can be connected to the internal network.61 2. most content and most application services were offered by commercial service providers. communities.Final Report: The Future of IP Interconnection 35 Web 2. but it supports the use of VoIP solutions because the open SIP standards are supported. and wikis. YouTube is a conspicuous example of this new development. Areas where these approaches might be reasonable and pertinent include intra-company communication. and the flow of information can be managed via search engines. for example. The content is shared among users. In the growth market of E-home. which information.0.

• • • • • In light of this evolution. Controlling of building services: Automation and remote control of home installations. Health: Monitoring of vital functions and automation of information exchange. (Tele-)Communication: Integration of all incoming and outgoing communication channels on different devices. office applications and hardware in the home and business environment. Presently. images. which classifies the relevant standards into different functional components. audio. . Leisure/entertainment: Sharing of entertainment applications in the home environment and with external users.36 Final Report: The Future of IP Interconnection • Integrated media consumption: Access to content (e.g. TV) from different sources via user-defined devices. security systems and domestic appliances. the provision of services at home is mainly based on different technologies. protocols and network infrastructure. it is clear that an extensive variety of technical devices may serve as parts of home networks. video. Figure 7 shows an E-home related protocol hierarchy. Home office: Sharing of data.

. … | Automatation | Sensors and Actors | | .GSM .. Voice . Jcontrol) . … | Mobile Phone...(Embedded) Linux . Video.MP3 ..MAC OS -… Virtual Machines . 63 See SerCHo Consortium (2005). Games) | Communication | PC. Text . Runtime Environment Operating Systems . XP MCE .. PDA.NET Device Platform | Entertainment (Audio. Source: SerCHo Consortium63 In the future. Game console..NET OSGi Communication Security Transport RTP UDP HTTP TCP IP Wireless UMTS GSM GPRS WLAN Bluetooth Wired DSL Ethernet Firewire LonMarks EIB/KNX Signaling SIP UDP Media Audio . . Settop-box..WMF -… Video . we anticipate a strong tendency toward convergence in order to foster interoperability of services and devices..AVI .Final Report: The Future of IP Interconnection 37 Figure 7: Protocol hierarchy of E-Home related standards SerCHo-System Services Basic Services Middleware HGI Jini HAVi UPnP DLNA MHP J2EE .LPCM .Java (J2ME.. The most important standardisation activities in the context of E-home focus on middleware aspects where actually still a lot of competing and overlapping standards are available.MPEG-2/-4 .Windows CE.Hypertext . Figure 8 illustrates the development of these standards over time...

OSGi is a component framework for service-deployment and lifecycle-management with a broad spread in other markets as well. which will in turn drive a strong need for interoperability and interconnection. nor a single unique value chain. thus enabling service deployment and lifecycle management as well. MHP. a child of the DVB consortium is a middleware solution designed to provide an abstraction of settop-box hardware and software. Both MHP and OSGi are device-independent since they rely on Java. There will be not a single unique business model. as illustrated in Figure 9. mainly in the computer and consumer industry. The complexity of E-home results not only from the multitude and variety of relevant standards. DLNA and HGI are important standardisation bodies. 64 See SerCHo Consortium (2005). As a consequence. DLNA and UPnP have a broad supporting community. many distinct entities involved in the development. the essential link between LAN and WAN. HAVi. .38 Final Report: The Future of IP Interconnection Figure 8: Timeline of E-home related standards 1999 2000 Auto-Config WG 2001 2002 2003 DSL Home 2004 2005 2006 Home API UPnP Forum UMA OSGi HGI HAVi DHWG DLNA SerCHo Source: SerCHo Consortium64 For network connectivity and device detection UPnP. deployment and provision of E-home based services. we anticipate a further increase of complexity in the field of E-home. Moreover. while HGI is mainly a network operator-driven initiative with a focus on the residential gateway.

the focus may be somewhat different than it has historically been the case in PSTN interconnection. The relevant issues relate to security. home networks)* Fully integrated E-home provider* * Possibly with integrated basic services. and the likelihood that services will be provided by third parties that are not necessarily the provider of network access to the house. and requirements regarding quality and security. Digital Rights Management (DRM). What is obvious is that the developments tackled bring about fundamental changes in particular regarding • • • physical communications infrastructure. . because many of the salient issues are more relevant to the application level. further services supplied by specialised content providers Source: wik-Consult In consideration of the complex structure of stakeholders in the technological process of providing home networks and associated services.3 Long term versus short term effects Many of the aforementioned changes are already under way even though the exploitation of their full potential still might need some time. traffic volumes and flows. and accounting. 2. At the same time. IP interconnection may be a critical success factor. transport network and platforms* Partly integrated manufacturer: management of platforms and user interface (devices.Final Report: The Future of IP Interconnection 39 Figure 9: E-home value chain and potential business models User interface Access Transport Platform Service Specialised service provider: management of platforms Partly integrated network provider: management of transport network and platforms* Fully integrated network provider: management of access network.

estimates of the rate of growth of IP-based traffic have varied enormously. and more recently even that growth rate has declined towards 50-60% per year. Short(er) term effects are e. the phasing out of MDFs.65 Given that many calls for changes in network interconnection payments in the IP-based world have been based on the assumption that traffic is about to explode. • In the access networks deployment of o o • • • • VDSL/FTTstreet cabinet technologies FTTB/H technologies Mid and long term effects are e. Andrew Odlyzko’s group at the University of Minnesota. …Traffic growth rates of 50% per year appear to only about offset technology advances. for a more or less long time there will be a co-existence of old networks (TDM) and new networks (access and core NGN).4 Implications for IP-based traffic patterns Over the past decade. Internet traffic growth appears to be decelerating. leading to a catastrophe for network operators. PoPs the migration to IP over Ethernet over fibre 2.40 Final Report: The Future of IP Interconnection The migration of traditional communications networks to new designs entails new topologies and architectures. there was a brief period of ‘Internet traffic doubling every 100 days’ back in 1995-96. but a number of groups have made credible estimates. It is not easy to develop reliable estimates of Internet growth.S. it is worth considering actual growth trends. Nonetheless. In the United States. when in fact it was doubling roughly once per year.g.g. they go on to say that “… although service providers are pushing to throttle customer traffic. but vastly different from what many in the industry had assumed. an argument can be made that they should instead be 65 Odlyzko (2003) explains the sources of rumors that Internet traffic was doubling every 100 days. find overall U.” Significantly. but already by 1997 growth subsided towards an approximate doubling every year …. Some estimates have been only weakly grounded in real data. as transmission capacity available for a given price steadily increases. a significant reduction of interconnection points . . and reaches the following conclusions: “In spite of the widespread claims of continuing and even accelerating growth rates. and global IP-based traffic growth rates of 50-60%. This is still an enormous growth rate.

their findings tend to contradict the notion Key Concept Box 9: Internet Traffic Growth Estimates of Internet traffic growth have varied widely.html. Telegeography found that “… [i]ntra-European Internet traffic grew 71 percent in 2007. that growth rate is about 50-60% per year. International Internet Traffic and Bandwidth Growth.” Like the Minnesota group. peak international Internet backbone traffic grew 60 percent between 2006 and 2007 (see Figure 1. and continues to decline. if then. Only during a brief period between 1995 and 1996 did Internet traffic “double every 100 days”. but huge by the standards of most other industries. On average. viewed on 14 December 2007. consumer IP traffic will bolster the overall IP growth rate so that it sustains a fairly steady growth rate through 2011. with various forms of video likely to play a large role going forward. page 4. Consequently.Final Report: The Future of IP Interconnection 41 encouraging more traffic and new applications. Currently. 68 See Cisco (2007). which corresponds to a quintupling from 2006 to 2011. Indeed. From 1997. that traffic growth is exceeding capacity growth: “In 2007. That's down from 74 percent growth in the previous year.dtc. which finds that “…international Internet traffic grew 57 percent from mid-2006 and mid-2007. Continued in Figure 11: Network Externalities on page 44.68 which predicts that “ … IP traffic will nearly double every two years through 2011.umn.edu/mints/home. . 67 Telegeograpy (2007). … Driven by high-definition video and high-speed broadband penetration. but will decline somewhat as a fraction of total Internet traffic. while bandwidth grew 68 percent.” These estimates are again consistent with a comprehensive study by Cisco Systems. worldwide growth subsided to a still impressive rate of doubling (plus or minus 30%) every year.”67 Some interviewees felt that IP traffic growth was much greater in Europe than in North America. One credible projection of consumer IP traffic projects a compound annual growth rate of 37 percent. to fill the growing capacity of transmission links. growing at a compound annual growth rate (CAGR) of 37 percent and nearly quintupling the monthly traffic run rate from 2006 to 2011. while VoIP traffic will continue to be small as a fraction of the total.” 66 See “Minnesota Internet Traffic Studies (MINTS)” at http://www. peak utilization declined somewhat. 2004-2007). compared with 85 percent the previous year.”66 These estimates are generally consistent with those of Telegeography. P2P will continue to grow. from 47 percent in 2006 to 44 percent in 2007. international Internet network capacity grew faster than the volume of traffic carried on these networks. The character of IP-based traffic is changing over time.

. Nonetheless. They expect healthy continuing growth for peer-topeer (P2P) traffic. In the near term. with various forms of video likely to play a large role going forward. They do not suggest a radical discontinuity with present trends. the Cisco study finds that “… [t]he Internet is not collapsing under the weight of streaming video. at rates of 35% per year for the next few years declining to 30% per year by 2001. these historical estimates and forward-looking predictions point to substantial ongoing IP-based traffic growth. P2P will decline somewhat as a fraction of total Internet traffic. as shown in Figure 10 below.” The character of Internet traffic is changing over time. but at rates that are either stable or gradually declining in percentage terms (albeit on a growing base).42 Final Report: The Future of IP Interconnection Significantly. the most formidable challenge that online video poses for the Internet will be flash crowds rather than the overall volume of traffic. Figure 10: Cisco estimates of global consumer Internet traffic Source: Cisco (2007) Taken together. Note that VoIP traffic will continue to be small in comparison to other services.

15:44 UTC. both traditional CPNP and newer negotiated “Coasian” arrangements. which is the topic of Chapter 4.1 Network effects (network externalities) Economists speak of network effects when the value of a network is a function of the number of parties who participate in the network. 'Externality'. mobile network and Internet arrangements are presented. It begins with key theoretical underpinnings: market power. and for the IP world.1. 3. PSTN.1. This is not a mere question of economies of scale. 21 May 2007.php?title=Externality&oldid=132458322> [accessed 25 May 2007] . an externality is “… a cost or benefit from an economic transaction that parties "external" to the transaction receive. and their implications for retail prices. It concludes with a lengthy assessment of welfare implications of different wholesale regulatory arrangements. while the second is the benefit derived by the party to a call who is not responsible for the call (often the call receiver). both at retail and at wholesale level. recurring themes in the economics of interconnection for the fixed (PSTN) and mobile (PLMN) world. the IP-based voice service. The Free Encyclopedia.wikipedia. 3.1. and overall consumer welfare. adoption.1 Externalities In economics. use. where unit costs decline as the number of units produced and 69 Wikipedia contributors. we first need to consider the implications of different levels of Quality of Service (QoS). Wikipedia. and transaction costs. These include network effects (a form or network externalities).Final Report: The Future of IP Interconnection 43 3 Economic theory and practice: Interconnection today and tomorrow This chapter deals with the economics of interconnection. and transaction costs. we discuss two externalities: the first (network externalities or network effects) relates to the benefits to all users of a network as the number of participants in the network increases. Before we can consider the economics of VoIP interconnection. especially in conjunction with the migration to Next Generation Networks (NGNs). <http://en. providing both theoretical and empirical assessments. 3.org/w/index.1 Key underlying economic principles This section introduces a number of common.”69 In this section. market power. Chapter 5 returns to economic analysis to consider economic implications of interconnection of the voice service as networks evolve to VoIP. network externalities.

it can be very difficult to get the market to move to a significantly superior (but less than perfectly compatible) service. The level or adoption often stalls at a level far less than the socially optimal level. would you have considered using email to invite your friends to a private party? For many of us. Moreover. Ten years ago. where an adequate service has been widely adopted.44 Final Report: The Future of IP Interconnection consumed increases. that act of joining enhances the value of the network to all network users. because there is one more person that they might conceivably contact. it enabled interaction with only a few thousand people. it might have far less value. the answer would have been no. because too few of our friends were reachable. Rohlfs . To an economist. Source: wik-Consult Let’s consider a concrete example. Your telephone service has enormous value to you because it potentially enables you to interact with any of billions of people. to get over the initial adoption hump. Today. or only a few dozen. Note that the shape of the curve in Figure 11 is irrelevant – what is important is that it slopes upward. however. it is linked to the number of parties with whom the subscriber could potentially interact. even though we ourselves had e-mail. if. rather. It is often difficult for a new service to reach the point where it serves a critical mass of people – that is. that the value perceived by each user increases with the number of participants. which makes the service much more valuable to you. even though they were not parties to the transaction. Figure 11: Network Externalities Perceived Value Network Externality: A new customer who joins a network enhances the value of the network to all network users. the answer is probably yes – the chances are that most of your friends can be reached by e-mail. that is. Positive Externality Number of participants Continued in Reciprocity vs. Symmetry on page 61. A new customer joins a network by engaging in a transaction with the service provider. these network effects are a form of network externality. however. Initial introduction of a service that is subject to network externalities poses special challenges.

Where network effects are in play. were in moderately wide use for time-shifting of television programs before a video rental business emerged. the amount of call externalities increases with the number of people who are on the network. the more recent discussion about termination monopolies has made it clear that. interconnection. Conversely. particularly in an IP context. the receiving party is subject to a call externality. 70 See Mitchell/Vogelsang (1991). pp. . call externalities were largely neglected in the telecommunications literature under the argument that calling and receiving party can internalize such benefits directly. however. In contrast. the externality is an external benefit to the caller. 81-83. Thus. Otherwise.” 3. But there are also many examples of seemingly meritorious products and services that never achieved sufficient deployment. for example. the importance of the call externalities relative to network externalities rises with the level of penetration. the question of which product or service will ultimately succeed in a market characterized by network externalities is not uniquely determined by the quality of the service.1. from both the players and the content. such internalization does not occur.2 Call Externalities Call externalities refer to benefits (or disutility in case of nuisance calls) received by a party to a call who is not part of the calling transaction with the network. Videocassette recorders (VCRs). “nothing succeeds like success. Moreover. if the caller is responsible for the transaction. 3. In general.1. success may instead be determined by which product or service is the first to secure a sufficient user base. This consumer surplus in turn usually is the sum of the benefits from being able to call (at certain prices) and from being called by others.1.2 Market power Market power is a central theme in European regulation. the more important call externalities become relative to network externalities. the called parties would choose to subscribe to networks that charge low termination charges. The European regulatory framework is largely premised on a traditional analysis of market power.Final Report: The Future of IP Interconnection 45 (2003) notes that different products and services have historically solved the initial adoption hump problem in different ways. It has been known for some time that call externalities and network externalities are interdependent. raises the threat of more subtle forms of exploitation of market power. The higher the penetration.70 The network externality arises from the expected consumer surplus enjoyed by joining a network. Historically. CDs appeared in part because there were companies (notably Phillips and Matsushita) that were positioned to profit from two vertically related markets. most of the time. if the receiving party is responsible. however.

3.1 Traditional market power In an ideal. firms would in effect be forced to accept the price that the market was willing to pay for their goods or services. Most markets are. or PATS) generally have market power relative to the termination of calls to their end-user customers – a phenomenon known as the termination monopoly. but consumers are short-changed to the extent that more goods and services could have been produced and consumed at a lower yet cost-covering price.73 We return to this point repeatedly.4. Where a company that has market power chooses to produce less than it otherwise could. If. either individually or jointly with others.2 Network effects and market power Where network effects are present. Article 14(2). Publicly Available Telephone Services72. The large player derives market power from its control over access to customers.1. then that large player may be motivated to implement less than optimal interconnection. one market player has a sufficiently large share of network customers. all market participants are likely to be motivated to provide near-perfect interconnection with one another in order to maximize the benefits of the network externalities. a special form of market power can come into play in connection with interconnection.g. perfectly competitive market. 71 See Directive 2002/21/EC. but were not. This loss to consumers is a deadweight social loss.1. however. notably in Section 4. Crèmer/Rey/Tirole (2000). 72 See Directive 2002/22/EC. 3. 73 See Katz/Shapiro (1985). These imperfections often enable firms to offer a smaller quantity of products or services than the market is capable of accepting. but to thus command a higher unit price (thanks to the normal working of supply and demand).2. the company may earn a good profit.”71 Incumbent providers of electronic communication services often have Significant Market Power as regards their last mile access capabilities..2. customers and ultimately consumers. This is so to the extent that no other service provider can complete calls to the telephone number associated with the end-user. imperfect to a greater or lesser degree. To the extent that market shares are well distributed over multiple market players. it enjoys a position equivalent to dominance. that is to say a position of economic strength affording it the power to behave to an appreciable extent independently of competitors. Article 2(c). however. It was only much later that . Farrell/Saloner (1985). Note that the earlier papers deal with interoperability in the context of standards compliance. For purposes of European regulation. Perfect interconnection would interfere with its ability to exploit its interconnection. a firm is “… deemed to have significant market power if. Providers of telephony services (e.46 Final Report: The Future of IP Interconnection 3. both in terms of the overall market and also relative to its next largest competitor.

which empowers “national regulatory authorities … to impose: (a) to the extent that is necessary to ensure end-to-end connectivity. See: http://www. Still. including IP version 6 (IPv6). inasmuch as any change has limited effect until it has been implemented for a substantial fraction of the interconnected networks. the same constellation of factors has tended to delay implementation of differentiated Quality of Service (QoS) between IP-based service providers. The most noteworthy mechanism appears in Article 5(1) of the Access and Interconnection Directive74. by some measures.5 of this report. each pair of networks that wish to directly exchange traffic need in general to execute a set of contractual arrangements. but has only limited mechanisms to deal with this subtler form of market power. .” In the context of interconnection. economists realized that network interconnection was essentially the same problem as standards compliance. 3. the largest IP backbones have tended to have dozens of interconnection (peering) agreements (not counting the transit services that they offer to customer networks) rather than hundreds or thousands. largely because it is seen as potentially being too powerful a tool and insufficiently constrained. but it has rarely been used. typically accompanied by a number of operational agreements and understandings. The combined effects of transaction costs and network effects have arguably played a large role in delaying the roll-out of any number of seemingly desirable Internet capabilities. but it also shows that the upper bound is large.76 Most relevant to this report. or to the Regional Internet Registries (RIRs) for assignment. We return to this point in section 4. These interact with network effects.iana.3. to multi-homed organisational users. when buying or selling a stock. In practice. thousands of IP-based networks. must pay a commission to their broker. For example. including in justified cases the obligation to interconnect …” This authority is directly relevant to interconnection. This number represents a very coarse upper bound on the number of independent IP networks. Given that there are. 75 More than 46.3 Transaction costs A transaction cost is “…is a cost incurred in making an economic exchange. 74 Directive 2002/19/EC. 76 See Marcus (2004a). most people. the number is large enough to imply significant transaction costs in evolving the system.1. and DNS security.75 this could potentially represent daunting transaction costs. that commission is a transaction cost of doing the stock deal.org/assignments/as-numbers.Final Report: The Future of IP Interconnection 47 The European regulatory framework is well equipped to deal with conventional forms of market power.000 Autonomous System numbers have been assigned to ISPs. obligations on undertakings that control access to end-users.

it could be perfectly normal and rational. however. possibly predation. Figure 12: Two-sided markets Two-Sided Platform $ Free-to-Air TV Broadcasting Seek attractive programs: more viewers add value for advertisers. for an average incremental price to be less than the corresponding average incremental cost would normally indicate some market failure. a singles bar might find it rational to offer free drinks to ladies (or to men) in order to ensure that both are present in adequate numbers. . the two-sided analysis is likely to provide additional insights only for markets whose two-sidedness implies that the structure of prices matters. a singles bar serves to bring interested males and females together. A two-sided market has some unique characteristics that distinguish it from a conventional market. Advertiser $ $ $ Advertiser $0 Advertiser Viewer Advertiser Source: wik-Consult Nearly any market could be analysed as a two-side market.48 Final Report: The Future of IP Interconnection 3.4 Two-sided markets In recent years. For example. in a two-sided market.77 It can be perfectly rational to subsidise one side of the market or the other. and not just the level of prices.1. Analogously. In a conventional market. however. who bear the full cost of the service (see Figure 12). the structure of prices often plays a crucial role in bringing the two sides of the market together. Notably. free-to-air television is distributed free to consumers in order to ensure an adequate audience for advertisers. so as to ensure adequate participation from both sides. economists have come to recognize a class of market players (platforms) who serve to bring two distinct sides of a market together. 77 See Rochet/Tirole (2004). A cable television operator serves to bring broadcasters and advertisers together with consumers.

Today. These fixed fees may be instead of usage-based fees.2 reviews wholesale arrangements. it is important to remember that voice telephony is likely to represent only a tiny fraction of the traffic on future integrated Next Generation Networks (NGNs).2 Traditional telephony interconnection models Much of the economic theory associated with interconnection has been developed for voice telephony in the Public Switched Telephone Network (PSTN) and the Public Land Mobile Network (PLMN).2. At the same time. and also that some of these payment systems have evolved significantly in recent years.2. This chapter applies established theory to the networks of today. We return to many of the same economic themes in Chapter 5. There may be higher charges for calls that travel a greater distance. and thus recover most of their costs. Most of us grow up in one system.1 reviews retail arrangements for voice telephony. In this context. particularly in light of the ongoing convergence of the PSTN/PLMN and the Internet. The established theory represents a useful point of departure for understanding the economics of IP-based interconnection. . or that cross national borders. from voice services. Whether this will continue to be the case is something of an open question. and take these charging arrangements for granted.Final Report: The Future of IP Interconnection 49 3. They often reflect the duration of the call. 78 The European Commission took action in 2007 to reduce excess roaming charges. 3. telecommunication companies derive most of their revenues. Section 3.1 Retail arrangements for voice telephony Retail plans for voice telephony can reflect a wide range of factors.78 There can also be fixed fees per unit time (for instance. and much less than many forms of data. There are usually quite substantial surcharges for calls placed or received when one is travelling outside of the country in which the call was placed (mobile roaming). in which case they might reflect the time of day or the day of the week on which the call was placed. The transmission of voice requires relatively little bandwidth – much less than video. where we consider the impact of the emergence of VoIP and of NGN on current arrangements. section 3. It is important to keep in mind that substantially different systems exist in different parts of the world. the retail payments are payments that an enduser (whether a residential consumer or a large enterprise user) makes to the telephony service provider.2. or they may be in addition to them. monthly charges). The wholesale payments are payments that one network operator makes to another.

the distance to the recipient. with no per-minute charges.1. They offered a nationwide flat-rate mobile telephone plan.50 Final Report: The Future of IP Interconnection 3.2. . independent of usage. and that flat rates are likely in consequence to prevail whenever the underlying usage-based costs to the service provider are not prohibitively high. see de Graba (2000) and Armstrong/Wright (2007). The American mathematician Andrew Odlyzko has argued that consumers generally prefer flat rate plans to usage-based. 83 See Odlyzko (2001).1. For example. More recent economic theory has called into question this traditional view of cost causation.81 A number of economists have suggested that it would be more appropriate if payment were split more evenly between the party placing the call and the party receiving the call. and whether the recipient’s telephone service is in a foreign country. many operators offer some form of freephone service (“800” numbers) where the called party pays. These can be useful for customer service numbers. Laffont and Tirole80 speak of the principle of receiver sovereignty – if the recipient did not perceive value in the call. The party that places the call pays a price reflecting the duration (number of minutes) of the call.83 Experience tends to support his prediction. 3.2. 80 See Jeon/Laffont/Tirole (2004). The program 79 As a noteworthy exception. hotel reservation services. Jeon.2 Flat rate. A particularly dramatic example was AT&T Wireless’s introduction in the United States of Digital OneRate in 1998.. 82 In addition to Jeon/Laffont/Tirole (2004) and Hermalin/Katz (2000).82 Alternative means of payment are often available where CPP arrangements are the norm. no distance-based charges.1 Calling Party Pays (CPP) The most common arrangements worldwide are referred to as Calling Party Pays. The recipient pays nothing. or wherever the service provider does not wish the cost of the call to represent an impediment to the use of the service. and should be viewed as the primary cost causer. 81 See also Hermalin/ Katz (2004). the recipient may incur substantial charges for receiving the call. banded flat rate (“buckets of minutes”) An alternative set of arrangements that have become increasingly popular in recent years are based on a fixed fee (or flat rate) per unit time (usually per month). the recipient could simply hang up. and no domestic roaming charges. or CPP. and possibly also reflecting the date and time on which the call was placed. where a mobile call recipient is roaming (traveling outside of the country to which the subscription pertains).79 These CPP arrangements implicitly assume that the party placing the call derives most of the benefit.

http://www. The per-minute charges are not intended to be used on a routine basis. Digital OneRate transformed the U. 84 See FCC (2006). to the extent that usage-based average incremental costs in the United States are primarily a function of air time. For example. Flat rate plans. mobile arrangements. they are set at high or punitive levels.com/shop/plans/. but they are markedly less attractive than their North American counterparts. and they therefore perceive no financial disincentive to making extensive use of the service. rather. or sometimes as non-linear prices.t-mobile. Customers appreciate the predictability and the relative simplicity of the plans.0). so as to enforce the need to upgrade to the next higher band as the customer’s usage increases over time. Economists refer to these hybrid plans as two-part tariffs. mobile market.85 In U. There are different bands. These hybrid “non-linear” prices are linear in a mathematical sense. in addition to some fee per minute (or perhaps some fee per minute when a pre-agreed number of minutes were exceeded). Flat rate plans for mobile telephones have also become widely available in Europe over the past few years. Plans of this type usually incorporate a per-minute charge for minutes in excess of those permitted in the band. a monthly usage fee might be assessed. but it would be a mistake to analyse a banded flat rate plan as if it were a simple two-part tariff. off-net calls to (other) mobile operators (which would normally be associated with high mobile termination fees) are usually excluded. these banded flat rate plans have been well accepted by consumers and industry.Final Report: The Future of IP Interconnection 51 was hugely successful. representing different numbers of total air minutes per month.84 Monthly subscription fees can serve as an exclusive pricing mechanism. the consumer will tend to think of the plan as being purely flat rate. The subscriber enjoys a true flat rate plan so long as he or she does not exceed a maximum number of minutes. whether banded or not. or they can be combined or mixed with usage-based fees. Consumers think of the plan as providing a zero marginal cost. tend to encourage use of the service. 86 See. forcing AT&T Wireless’s competitors to merge or form alliances in order to achieve nationwide coverage so as to be able to package together a competitive offering.S. In the United States. . far beyond their expectations. They probably track underlying costs of mobile telephone service reasonably well.S. but the Y-intercept is not at the origin (0. So long as the consumer does not exceed the maximum number of minutes in the band. for instance. 85 This expression can be confusing to those of us who are not economists. it is common to find a variant of the flat rate plan that is sometimes referred to as the “buckets of minutes” plan.86 These plans are most appropriately viewed as representing banded flat rate arrangements. Notably.

and from the fact that consumers tend not to use the maximum number of minutes that they have purchased.2. Service providers benefit from a predictable revenue stream. or else they are voluntarily negotiated between the networks. agree not to charge one another (an arrangement known as Bill and Keep).1 below). however. 3. in 1998. and the receiving party also pays.87 Ten years ago. these arrangements have basically disappeared in North America. the calling party pays based on the number of minutes of use.2. . Starting with the introduction of banded flat rate plans in the U.3 Receiving Party Pays (RPP) and hybrid arrangements A few countries historically made limited use of an alternative set of arrangements known as Receiving Party Pays (RPP) at retail level.2. other things being equal.1. the consumer preference for banded flat rate has been overwhelming.2 Wholesale arrangements among switched fixed (PSTN) and mobile (PLMN) telephone networks Wholesale arrangements in the PSTN/PLMN generally are of one or two forms: either they follow the principle of Calling Party’s Network Pays (see Section 3. the fixed networks in North America have evolved toward flat rate plans. In an RPP system.1 discusses CPNP arrangements.S. A commonality between RPP and banded flat rate plans for mobile phones is that both generate charges that vary primarily with the total amount of air time used. the networks often.2 discusses voluntarily negotiated arrangements. Canada. more recently. both reflect both outgoing and incoming calls. 87 Europeans pay for calls received only in the case where they are roaming in another country.2.52 Final Report: The Future of IP Interconnection 3. these arrangements were the norm for the mobile industry in the United States. but not always. Section 3.2. If negotiated. and Singapore.2. Consumers tend to prefer banded flat rate because of its predictability. One would expect the price of placing a call in a RPP system to be similar to that of receiving a call. it is difficult to make a comparison inasmuch as no country that we are aware of ever implemented RPP for all calls. CPP was the norm for calls from fixed telephones in countries where RPP was used for mobile. In practice.2. In other words.2. while Section 3. Historically. and both prices to be about half of the price of placing a call in a CPP system.

2. and they are related to one another. the terminating network would not be paid at all for the call.Final Report: The Future of IP Interconnection 53 3. call termination payments made and received are comparable in magnitude. First. the CPNP wholesale payment is to a significant degree a reflection of the lack of payment by the call recipient under CPP – in the absence of a retail payment. 88 For some networks. Figure 13: Calling Party's Network Pays (CPNP) wholesale arrangements Call placed Call received Retail CPP Payment Originating Network Terminating Network Wholesale CPNP Payment Source: wik-Consult CPNP wholesale payments and CPP retail payments are often found together. The network shown as Terminating in Figure 13 is. including the relative number of calls originated and terminated between the two networks and the relative magnitudes of the termination fees per minute. the lack of a termination payment need not pose a problem.1 Calling Party’s Network Pays (CPNP) Calling Party’s Network Pays (CPNP) arrangements reflect a payment at the wholesale level from the originating network (the network whose customer placed the call) to the terminating network (the network whose customer received the call). . Whether the CPNP termination payments balance for a particular network depends on a number of factors.2. Between fixed networks.88 The CPNP wholesale payment thus seeks to correct for the lack of a retail payment from the party that receives the call. but they are not the same thing. of course. or between mobile networks. Figure 13 depicts the relationships and the flow of payments. Mobile termination fees tend to be much higher than fixed – nine times higher in Europe. according to the 12th Implementation Report – so the termination fees do in fact influence the relative positions of fixed and mobile networks. paid by its own customers for calls that they place. and thus have little direct influence on profitability.

91 In the interest of brevity. and would thus suppress demand to place calls to its customers. the smaller the market share of the service provider. in the absence of regulatory constraints. the higher the termination price that it is likely to be motivated to set. One must take care in comparing the figures. be motivated to set a high termination price. One of these is the terminating monopoly.2054 in 2001. many years for such solutions to become widespread. implies changes to programming and possibly the hardware of the installed base of GSM phones.3. it would take many. CPNP fees for fixed incumbents were generally set to cost-based regulated levels of less than € 0.1. Prior to the implementation of the current European regulatory framework in 2003. and also greatly in excess of real usage-based average incremental costs.92 The mobile fees were greatly in excess of fixed. but mobile termination fees were set by mobile operators without regulation and were on average in excess of € 0. that end-user is typically accessible only by means of that telephone service provider. 10th Implementation Report. the end-user usually has little or no visibility into the termination prices to individual networks. Once an end-user has selected a telephone service provider. A provider with a large market share would nonetheless be somewhat constrained by the likelihood that setting a high termination fee would cause its competitors to set even higher retail prices. prior to implementation of the current European regulatory framework (cf. Perversely. Kruse (2007). could potentially provide a very comprehensive way around the termination monopoly. One could imagine CPNP variants that would work a bit better than the current system. A provider with a small market share has a correspondingly lesser impact on overall retail prices.02 per minute of use. of course. however. . most notably because the size of the EU was different at the time of the 10th Implementation Report. for instance. 92 Fixed-to-mobile termination rates were € 0. this.2.90 Even where the market for call origination is effectively competitive. this discussion glosses over a number of assumptions that usually hold in CPP/CPNP systems. 89 The end-user might.1140 (cf. however. however. 12 Implementation Report. page 35). The service provider will. have different telephone services with different service providers and different telephone numbers. page 29). 90 In the longer term. a call placed to a particular telephone number can in general only be terminated by a single telephone service provider. For example. it is a different network provider’s customer whose demand is being suppressed.91 Experience in the European Union tends to confirm these predictions. Even a monopoly price is normally constrained by the risk that high prices will suppress demand. which provides a ranked list of different IPbased services that can be used to reach an end-user based on his or her telephone number. in this case.89 This effectively confers a form of market power on the terminating service provider. there will usually be market power in the market for call termination. Clearly. Annex 2. has suggested that the GSM standard might be modified to enable a preferred operator to complete a call to any phone. Annex 2. the th corresponding figure is just over half of that. and is therefore less constrained.54 Final Report: The Future of IP Interconnection A second relationship between the two is that the CPNP wholesale payment tends in effect to set a floor for the retail price of the call. however. CPNP suffers from a number of well-documented deficiencies. ENUM. or € 0. technical solutions to the termination monopoly might be possible.20 per minute of use. In 2007. We return to this point in Section 3.

These market power findings have empowered NRAs to impose remedies so as to address high termination fees.93 and nearly all have found it necessary to do so. Call termination on individual public telephone networks provided at a fixed location. due not only to new Member States joining the EU.96 but are probably still well above true average incremental costs in most if not all Member States. Note that figures are not strictly cross-comparable between the earlier and later reports.94 are: [Market] 3. EU-Commission (2007b). …95 The market definitions for termination of fixed and mobile calls are in terms of individual networks. 93 94 95 96 See Recommendation on Relevant Markets (2003/311/EC). See ibid. call termination is taken to include call conveyance. Voice call termination on individual mobile networks. 24-25 and 29. … [Market] 7. and also EU-Commission (2007). and Figure 16) . in the context of the analysis of a single network. See EU-Commission (2004).Final Report: The Future of IP Interconnection 55 In response to the termination monopoly. reflecting the updated market definitions issued by the Commission in December 2007. pp. . For the purposes of this Recommendation. the European Commission has so defined the markets for termination of fixed and mobile calls as to enable national regulatory authorities to regulate termination fees. Termination fees have been on a downward glide path since 2003 (as shown in Figure 14.Figure 15. with the delineated boundaries for the market for call origination and the market for call transit on the public telephone network provided at a fixed location. but also due to changes in the data reported. delineated in such a way as to be consistent. The pertinent definitions of wholesale termination markets susceptible to ex ante regulation. that a network operator will in general be found to have market power relative to its own customers unless there is some specific reason to be believe otherwise. in a national context. This means.

2001-2004 Source: 10th Implementation Report. . Annex 2.56 Final Report: The Future of IP Interconnection Figure 14: Fixed-to-fixed interconnection fees in the EU15. Annex 2. Figure 15: Fixed-to-mobile interconnection fees in the EU15. 2000-2004 Source: 10th Implementation Report.

3. or for the time of day or day of the week (e. Annex 2. In effect. for the distance that the traffic must be carried. presumably reflecting different average incremental costs for the respective networks. .g. Worse. inasmuch as it implies the need to maintain intensive regulation indefinitely. Substantially different rates have been set for fixed versus mobile networks.2. this outcome must be viewed as being less than ideal.2.1 Traditional reliance on minutes of use. CPNP arrangements have evolved somewhat over time in response to the realization that the average incremental costs of today’s communication networks are only weakly linked to these indicators (and that they probably will be even less closely linked to the costs of the Next Generation Networks that are about to be deployed). 2004-2006.1.Final Report: The Future of IP Interconnection 57 Figure 16: Fixed-to-mobile interconnection fees in the EU25. these were assumed to reflect the underlying cost of the network. the regulated call termination prices are still well above efficient levels that would maximize social welfare. Even though the Commission has responded effectively and appropriately to the problems associated with the terminating monopoly. termination fees have primarily been set based on minutes of use. Termination fees have sometimes reflected different rates for national versus local traffic. peak versus non-peak hours). and distance Historically. Source: 12th Implementation Report.

while double-transit termination would require PoIs only at remote switches. As a result.58 Final Report: The Future of IP Interconnection 3. single-transit and double-transit calls. but rather depend on the parts of the network that are required for a call. This prediction can be quite complex. Moreover. As there can be no doubt that most network costs depend on the capacity utilized. for which the maximal capacity utilization is booked in advance and paid in monthly or one-time fees.2. and because this dependency of costs on capacity is increasing rather than decreasing. Furthermore. What distinguishes CBC from per minute charges is the tracking of network costs and the possibility for risk sharing between the dominant network operator and the other competitors. Typically.2. Also. A major feature of EBC is that minutes are the primary measure of usage. capacity-based charging would appear to follow naturally in a CPNP environment. much more so than tracking types of calls through the network.2. because network costs are becoming progressively less distance dependent over time. EBC has been an efficiency improvement. EBC usually does not track peak behaviour very well. without triggering any further charges for usage as long as usage remains within the specified capacity limit. CBC therefore generally can follow efficiency criteria more closely than per-minute charges. types of calls are defined by sets of network elements that they use. This was quite adequate in the PSTN world. and double-transit categories by the Points of Interconnection (PoI) that are being used by the interconnecting parties. packet switching does not lend itself to minutes as a measure of network usage. Within the CPNP approach.2.1. A difficulty with CBC arises from the contractual arrangements between access seeker and network provider (the terminating network in the case of CPNP). Each type is characterized by a profile of network usage leading to characteristic quantities of network elements. peak usage patterns of all network users have to be . Calls are assigned to the local. but it does not carry over well into the IP world.2 Element Based Charging (EBC) Element Based Charging (EBC) has largely replaced distance-related charging of origination and termination services. it is necessary to establish which parts of the network will be used by interconnection and to what extent the interconnection requires additional capacity. In Europe. In order to conclude an agreement.3 Capacity Based Charging (CBC) Capacity Based Charging (CBC) relates to interconnect services. singletransit.1. The costs of these quantities are determined in bottom-up analytical cost models. where networks and services are sometimes distinct. local interconnection requires branching out the network much more than double transit. A network ‘A’ wanting to terminate calls on network ‘B’ using local termination would have to terminate in one of the many local exchanges of network ‘B’. 3. it is customary to distinguish between local.

Coase’s ideas have since been put into practice in many leading economies. Coase argued that private parties were generally much better positioned than regulators to negotiate mutually acceptable arrangements based on commercial principles. and even with the best of tools. provided that certain preconditions were met. Today. as previously noted. These negotiated arrangements can be viewed as generally reflecting the thinking of the Nobel-prize-winning economist Ronald Coase. In the case of spectrum. In Europe. Canada. In principle. the time frame for CBC has to be established. this has led to lower and arguably more appropriate termination rates than existed prior to 2003. it is difficult for a regulator to determine a rate that is altogether satisfactory. the arrangements can hardly be viewed as ideal. resulting in more risk sharing the longer the time frame. one would expect that Coase’s analysis would be equally applicable to PSTN/PLMN interconnection arrangements. 3. “The Federal Communications Commission”97.2. many interconnection fees are set to nominally cost-based rates by regulators. the Bell System had simply refused to 97 See Coase (1959). a key precondition was that “property rights” be adequately defined. He proposed that blocks of spectrum be managed more nearly like property. Most experts would regard spectrum auctions and spectrum trading as representing best practice today. will networks in fact agree to interconnect? What preconditions. rather than being assigned by (hopefully) benevolent bureaucrats as was then standard practice. One might expect that the network operators were better positioned to determine the right rates than is the regulator. In Coase’s famous 1959 paper. are necessary to ensure that a Coasian system would work for switched PSTN/PLMN network interconnection? What is the likely outcome of such arrangements? • The willingness to interconnect cannot be taken for granted. In the United States prior to the Kingsbury Commitments of 1912. Last.2 Negotiated “Coasian” arrangements The main alternative to CPNP arrangements is permitting network operators to negotiate fees among themselves. if any. in the form of spectrum auctions and spectrum trading. This line of reasoning raises several key questions.2. and Singapore): • • Under Coasian arrangements. however. Even with the best of intentions. which are answered in part by experience in the countries that have Coasian “Bill and Keep” arrangements today (notably the United States. .Final Report: The Future of IP Interconnection 59 aggregated.

law require Bill and Keep arrangements. For the PSTN. see Marcus (2004b). (Reciprocal compensation fees are two-way termination fees. but there is no obligation that aAC be equal to aBC. B and C) are operating in a Member State. The question of preconditions is complex. United States law explicitly permits network operators to mutually agree to pay no reciprocal compensation fees (a system referred to in the U. regulation has not been necessary to date. termination rates. which are one-way payments from the long distance network operator to the local network operator. Later in this chapter.. generally applicable to local interconnection. however. including zero. For a more expansive assessment of call termination arrangements in the United States. we consider why the experience has been so different. 98 See Mah (2003). the same in both directions. It is worth noting that when we speak of reciprocity in U. i. mobile operators and non-incumbent fixed operators usually choose Bill and Keep arrangements with one another. so long as the rates are reciprocal. this implies the need for some kind of regulation. we mean something quite different from the European notion of symmetric termination rates. Long-distance-to-local interconnection is implemented by access fees. if three mobile operators (call them A. as Bill and Keep). which is tantamount to setting the termination rate in both directions to zero. then symmetry means that aAC (the termination rate from A to C) must be the same as aBC. Nonetheless. however.99 Experience suggests that large PSTN incumbents with market power are unlikely to voluntarily interconnect with much smaller competitors in the absence of government intervention. all operators have interconnection obligations. are free to set any rate they wish. In the U. reciprocity obligations mean that aAB must equal aBA. In the United States. in the case of the Internet (and presumably also for other IP-based services).100) Wireless operators. In Europe. 99 See Haucap/Marcus (2005) 100 This discussion necessarily abstracts and simplifies what in reality is a very complex system.98 More recently.S.60 Final Report: The Future of IP Interconnection interconnect with competitors.S. and also non-dominant fixed operators.S. . in no case does U. but only fixed incumbents are subject to cost-based reciprocal compensation termination fees.e. Telecom New Zealand refused to interconnect with Clear for years until forced to do so by court action.S.

for networks A. Continued in Quality of Service economics and engineering on page 79. independent of the relative sizes of the networks. the payments that they make to other network operators. framed within a requirement to maintain reciprocity. the mobile operator cannot demand a higher termination fee than the fixed incumbent itself is permitted to charge (based on 101 This is the general expectation. It is not always the case in practice. but in addition their termination fees are capped by regulation so as not to exceed cost. including payments to mobile network operators. is a market-based outcome in a regulatory environment that bounds the negotiated options.103 In the case of the fixed incumbent.Final Report: The Future of IP Interconnection 61 This is a subtle but important point: The use of Bill and Keep arrangements is the U. they still influence retail prices. In other words. operators should set these rates to monopoly levels. Without reciprocity. . In effect. the obligations of symmetry appear to play a central role in motivating the mobile operators to agree to a Bill and Keep system in regard to these two-way reciprocal compensation arrangements.1. unlike most CPNP regimes. but rather by the absence of an Key Concept Box 10: Reciprocity vs. reciprocity mean that aAB (the termination rate from A to B) must equal aBA. reciprocal compensation system is characterized. the dominant fixed incumbent usually interconnects under arrangements that include CPNP-like termination fees at the capped rate. Even though the payments net to zero. two interconnected networks charge each other the same termination rate per minute. affirmative obligation to pay something other than zero.S. moreover.S. however. 103 See Armstrong/Wright (2007). however. With this in mind.3. change the negotiating game in a fundamental way – they prevent the mobile operator from collecting highly asymmetric payments from the fixed operator. Symmetry With reciprocity. Mobile-to-mobile (M2M) traffic will tend to be roughly balanced in any case.101 so there is typically no reason for the mobile operators to prefer a high termination fee to a low one. The U. Symmetry means that aAC must be the same as aBC. We return to this point in section 3. 102 See also Laffont/Tirole (2000).102 The symmetry obligations between fixed and mobile. Fixed incumbents are subject to the same obligations. or vice versa – the charges will tend to net to zero whether the fee is high or low. and B. are at the same respective rate for each corresponding network as those that the fixed incumbent collects. a network charges the same termination rate per minute to each network with which it is interconnected. not by a regulatory obligation to charge nothing for call termination.2. which independently models mobile-to-mobile and fixed-to-mobile termination arrangements. With symmetry.

the termination fee to and from fixed incumbents is effectively capped at rates well under €0. the long distance network operators were historically distinct from the wired incumbents. there was only a single network that they could use. exchange traffic pursuant to negotiated arrangements. the regulator need not be involved with setting termination rates.104 Long distance network operators were obliged to make oneway payments (access charges) to local network operators. This option has rarely if ever been exercised.) The FCC found it necessary. and mobile usage is consequently much higher than in Europe. As we explain later in this chapter. the average normalized price of mobile services. see Rohlfs (2007). 104 A series of mergers over the past few years have blurred these distinctions today. As in the United States. low termination rates lead to low retail prices and high usage.62 Final Report: The Future of IP Interconnection the costs of the fixed incumbent). mobile operators are not entitled to collect access charges at all. enormously in excess of their cost. except for those of the fixed incumbents. Significantly. The dominant fixed network operator interconnects with other networks pursuant to CPNP arrangements at regulated rates. and unless the mobile operator expects to terminate far more calls than it originates it has no reason to prefer a charge at the high end of the range over a charge of zero. Fixed-to-mobile and mobile-to fixed charges are low. the negotiated arrangements usually lead to an agreement to not charge termination fees. law and regulation. Again. Mobile operators. Rohlfs finds that countervailing buying power cannot ameliorate the terminating monopoly. Canada and Singapore are working well.106 Singapore’s arrangements are surprisingly similar to those of the United States. Thus.S.107 The arrangements in the United States. There are no charges for mobile-tomobile termination.e. to impose new regulations to limit the access charges of small (non-dominant) fixed operators – they were unilaterally imposing fees of several cents per minute. measured in terms of service-based revenues per minute of use. but it is slightly different.01 per minute in any case. 106 They also invited operators with higher costs to document their costs in order to receive higher access fees. known as Regional Bell Operating Companies (RBOCs). . however. but they are still reflected in U. on the order of half a cent per minute. In the United States. completing the call. 105 For an analysis of countervailing buying power and termination fees.005 per minute. (Curiously.105 The FCC solved the problem by limiting these non-dominant fixed operators to the same fees charged by the incumbent operator for the same area. i. and non-dominant fixed operators. are generally limited by regulation to access charges of less than € 0. and in the absence of the ability to charge that network they had no countervailing bargaining power. to a Bill and Keep system. are low in these systems. The former Bell system incumbents. This can be viewed as a variant of the termination monopoly problem – the long distance network operators were obliged to complete the calls. and arguably offer advantages in comparison with current European practice. 107 Hong Kong’s system is also interesting.

the originating network might choose to disregard it and offer a flat rate at retail. 111 See Laffont/Tirole (2000). but the access charge affects each network’s perceived average incremental cost and therefore retail prices. it usually leads to an even higher CPP retail price per minute. pp 201-202. 112 It is unusual.02 per minute). because they do not incur termination charges in this case. but it is not unheard of. Flat rate plans typically exclude the ability to call phone numbers associated with high CPNP wholesale fees. For on-net calls – calls from one subscriber of a network to another subscriber of the same network – operators can and often do offer lower prices that correspond to the operator’s real costs. The willingness of network operators to charge less for on-net calls than for off-net calls strongly suggests that the network operators themselves view termination fees as being well in excess of the real costs that they themselves incur. the associated monthly charges are usually so high as to be unattractive to most potential subscribers. 3. are often too high to ignore – they sometimes represent the preponderance of the total cost of the call. independent of the relative sizes of the networks. more than about € 0. for adoption of services. however.112 even where termination payments are likely to net to zero. 108 Where flat-rate plans are available that include calls to networks that collect high termination fees. Network operators will tend to be reluctant to offer service at a average incremental price below their average incremental cost.2. .113 If an operator were to set its retail price for off-net calls below the level of the termination fee. It is not always the case in practice 110 Laffont/Tirole (2000).2. as Laffont and Tirole have noted: “It is correct that a change in the access charge need not affect the (absence of) net payment between the operators.”110 Each network operator views its payments to other network operators as a real cost. Where the CPNP payment is small.3. therefore. however. an operator might choose to ignore a termination fee that constitutes only a small fraction of the total cost of the call. and for use of the service once adopted. Mobile termination fees set in the absence of regulation.Final Report: The Future of IP Interconnection 63 3.1 Relationship between wholesale arrangements and retail arrangements The CPNP wholesale payment tends in effect to set a floor for the retail price of the call. 109 This is the usual expectation. however. It is. it is unusual to see retail prices below a “high” wholesale call termination rate.111 For off-net calls (calls to a subscriber of another network).109 and if they were to charge identical termination fees to one another. 113 Again. where the CPNP wholesale payment is large (as a rule of thumb.3 Implications The use of CPNP versus Coasian arrangements has significant implications for retail price. one might be tempted to think that termination fees do not matter very much. then there would be a zero net payment between them. Since termination fees have no impact on net payments under these conditions. not neutral even if traffic is balanced. however.108 If traffic were balanced between two operators. This is true whether the termination fees are low or high.

3. expressed in US dollars.2. The original source of the data is a series of well-regarded quarterly surveys by MerrillLynch. each at a price below cost. and CPNP The implications of termination arrangements on consumer welfare are quite complex. but also the actual use that is made of them. Note that the y axis of the graph reflects a logarithmic scale. selected data are cited annually in reports from the U. (2) services associated with high CPNP termination fees are typically excluded from flat rate retail plans. 115 Service-based revenue excludes equipment. These data correspond to the last quarter of 2005.S.2 Demand elasticity. These tendencies are much more visible in mobile services than in fixed. thus generating a net payment (an access deficit) to other operators. for reasons that we return to shortly. .115 114 See EU-Commission (2007). Comparative international data suggest that (1) high CPNP termination fees lead to high retail price per minute. in comparison with the total number of MoUs for mobile services in a number of leading economies around the world.64 Final Report: The Future of IP Interconnection it runs the risk that its customers will capitalize on the favourable price by placing too many off-net calls.114 Figure 17 shows the relationship between the service-based revenue per minute of use (MoU). and (3) end-users consume less of services that have high prices. retail price per service minute. and thus does not take handset subsidies into account. Servicebased revenue per minute is a useful normalized proxy for retail price – it has the advantage that it already reflects not only the plans that are available. There is a slight anomaly where Merrill-Lynch correctly counts minutes for Bill and Keep countries. because mobile termination fees are much higher than fixed – about nine times higher in Europe.3. FCC. but counts on-net minutes in CPNP countries only once rather than twice – this does not change the conclusions that follow.

Final Report: The Future of IP Interconnection

65

Figure 17: Mobile revenue per MoU versus MoUs per month

Mobile Revenue per MoU versus MoU per Month
1000 US, 798

Minutes of Use (MoU) (logarithmic scale)

Hong Kong, 395 Singapore, 313

Canada, 403 South Korea, 322 Finland, 279 France, 235 Australia, 178 Sweden, 141 UK, 146 Spain, 150 Italy, 126 y = 685.2e-7.1487x R2 = 0.7873 Japan, 147

100 Germany, 81

10 $-

$0.05

$0.10

$0.15

$0.20

$0.25

$0.30

Service-based Revenue per Minute of Use (MoU)

Source:

wik-Consult, based on data from the US FCC’s Annual Report and Analysis of Competitive Market Conditions With Respect to Commercial Mobile Services, 11th Report (11th CMRS Competition Report), September 2006, Table 12, based in turn on Merrill Lynch Telecom Services Research, Interactive Global Wireless Matrix 4Q05.

Countries where service-based revenue per MoU is high tend to have low usage – for example, Germany (with service-based revenue of $0.28 per MoU) has just 81 minutes of use per month. This suggests that prices are so high that end-users are reluctant to use their mobile phones. Conversely, countries where service-based revenue per MoU is low tend to enjoy high usage. Singapore (with service-based revenue of $0.08 per MoU) has a substantial 313 minutes of use per month. The United States (with service-based revenue of $0.07 per MoU) has a whopping 798 minutes of use per month. It is possible that US consumers are responding not only to low prices overall, but also to the prevalence of flat rate arrangements – they tend to perceive the average incremental cost of each additional phone call as nil, and are therefore willing to use their mobile phones with abandon. Economists speak of the price elasticity of demand. Where prices are high, consumer demand is suppressed. Where prices are low, consumer demand is stimulated. The line on the graph represents a best-fit (regression) exponential trend line, and might be viewed as an estimate of this demand elasticity. This trend line appears to be linear because the data have been plotted on a logarithmic scale. The trend line is associated with an R2 value of 0.787, a goodness of fit measure that suggests that the trend line fits

66

Final Report: The Future of IP Interconnection

the data well. At the same time, one should be careful about interpreting this data as elasticity – consumers in Germany cannot really substitute a U.S. mobile service for a German mobile service, and Japanese consumers may be either more or less talkative than Italians for cultural reasons.116 Nonetheless, the data strongly suggest that demand is elastic, which is to say that demand responds strongly to price.117 Companies seek to maximize profits. Other things being equal, they will seek to maximize gross revenue as well. The Average Revenue per User (ARPU) is a useful measure of the contribution of each end-user to that gross revenue. In this case, high prices are not necessarily indicative of high ARPU. To understand this, consider that the (service-based monthly) ARPU can be viewed as being the product of the service-based revenues per MoU and the MoUs per month.118 Germany’s service-based revenues per MoU are four times higher than those of the US, and nearly twice as high as those of France; however, U.S. usage is ten times higher than that of Germany, while French usage is three times higher than that of Germany. Germany’s service-based revenue is four times higher than that of the US, but U.S. (service-based) ARPU is nonetheless 2.5 times higher than that of Germany. Germany’s service-based revenue per MoU is nearly twice as great as that of France, but French ARPU is nonetheless nearly twice that of Germany.119 This is a striking observation – it suggests that European mobile operators in many cases could actually increase ARPU by lowering prices.120 Moreover, to the extent that they are maintaining high prices to enhance profits at the cost of suppressing use, the suppressed usage should be viewed in economic terms as a deadweight social loss, a loss of consumer welfare. In other words, these high prices clearly harm consumers, and are so high as to also reduce European gross revenues and possibly in some cases the profits of the firms themselves. It is also worth noting that high retail unit prices do not necessarily correspond to higher unit costs. Other things being equal, one would expect high density countries to have lower unit costs for mobile service than low density countries; however, the U.S., Canada and Australia (all relatively low density countries) all have lower service-based

116 If mobile roaming charges were negligible, perhaps they could. 117 In fact, this simple graph suggests that demand might be far more elastic than most economists have assumed. 118 In general, for any point on the graph, a country’s ARPU is the area from the (0,0) point to the point associated with the country. In light of the logarithmic scale, however, one cannot simply compare areas. 119 The Merrill-Lynch data are expressed in U.S. dollars at then-pertinent exchange rates. The comparison would look a little different at today’s exchange rates, but not enough to change the conclusions. 120 In the rather extreme case of Germany, the unusually low usage might suggest the presence of double marginalisation. The combined effects of high wholesale termination rates and even higher retail prices have an effect equivalent to that of multiple monopolists in vertically related markets, and collectively drive usage levels down even lower than the level of monopoly pricing (i.e. below the levels that are profitable for the firms themselves). We have not assessed the degree to which this might be present in other Member States.

Final Report: The Future of IP Interconnection

67

revenues per MoU than do Japan and Germany (both of which have high density). Nor could technological differences have driven lower unit costs – the U.S. market is about 45% comprised of GSM, at frequencies not so different from those in use in Europe. The CPNP arrangements that are universally in place in the European Union are widely accepted, but they are not without cost. To the extent that they lead to higher unit prices at retail, and thus to suppressed use of services, they negatively impact consumers.

3.2.3.3 CPNP and mobile adoption Most experts feel that CPNP, particularly when associated with relatively high termination fees, tends to lead to rapid take-up of mobile services.121 Mobile operators are motivated to offer plans (especially pre-paid plans) with no initial fee and no monthly subscription fee, and to subsidize the cost of the end-user’s handset, largely because of termination fees that are well in excess of cost. Even if the customer makes few (or no) outgoing calls, the calls that the end-user will receive make it profitable to gain him or her as a customer. Bomsel et al. (2003) found that high fixed-to-mobile (F2M) termination rates resulted in the involuntary transfer of many billions of euro per year from fixed customers to mobile operators.122 Fixed operators rarely complain, because their own mobile operations benefit; however, the system as a whole may have an overall tendency to distort the evolution of network, favouring the development of mobile networks over that of fixed to a degree not otherwise warranted.123 Mobile operators may have some tendency to compete away some of these profits through promotional handset subsidies and other promotional incentives, but they retain more than they return to consumers.124 Promoting penetration of Publicly Available Telephone Services (PATS) is a meritorious goal in general, but it is not clear that promoting mobile penetration at the possible expense of fixed is an appropriate policy goal for the EU at this point in time. According to the 12th Implementation Report, mobile penetration in the EU25 as of October 2006 was 103.2%, having increased 8.2 percentage points from October 2005. This is to say that the number of mobile subscriptions was greater than the number of inhabitants. Seventeen of the twenty-five then-current Member States had penetration rates in excess of 100%. Only France (82%), Malta (83%) and Slovakia (86%) had penetration less than 90%.125

121 This is the prevailing wisdom, but it is not universally accepted. See for instance Littlechild (2006) and Dewenter/ Haucap (2005). 122 WIK estimates the financial transfers from fixed to mobile networks in Germany to have been about 10 ten billion (10 ) euro for the period 1998-2006. 123 For example, these transfers may lead to a faster decline of fixed network subscriptions and of fixed network usage than would otherwise be the case. They may be a significant factor in the dramatically rapid decline of fixed subscriptions in some Member States. 124 See also Littlechild (2006) and also Genakos/Valletti (2007). 125 See EU-Commission (2007), pp. 37-38.

” It drives the termination rate too high. Historically. 3. 128 Page 18. In fact.3. computing this cost involves numerous complexities. A Eurobarometer survey found that 20% of Europeans do not have access to a mobile phone. and possibly a range of other effects as well.126 Nonetheless. to be sure. but also that there never will be. mobile penetration is clearly high. and growing. including the degree of competition in the fixed telecommunications market. and the pattern of demand of mobile subscribers.4 What is the “right” call termination rate? Determining the “correct” termination rate is not at all straightforward. an imperfect measure of access to PATS. particularly at a time when the fixed network appears to be critical for the next phase of broadband deployment. they argue that “… including fixed and common costs [as a markup to regulated termination charges] is likely to be a flawed policy. Economists.” 126 Eurobarometer. the type and intensity of competition in the mobile outbound market. have felt for quite some time that an optimal termination rate would also take into account network externality benefits on the part of the recipient.2. who is quoted in this section. There are also a number of other factors that could affect the welfare maximizing charge. . It is not clear that it is appropriate to provide further stimulus to the mobile network at the expense of the fixed network. one leading expert127 has argued cogently not only that there is no consensus on the proper theoretical basis for the termination rate. In practice. 127 Littlechild (2006). the existence and significance of network externalities. …[T]he direct intervention to set the welfare maximizing fixed to mobile termination charge requires a regulator to form a view on the demand elasticities. Armstrong and Wright (2007) find that the termination rate that optimizes welfare is less than marginal cost by the value of the network externality to the recipient of the call.68 Final Report: The Future of IP Interconnection Penetration as routinely measured (dividing the number of subscriptions by the number of inhabitants) is. E-Communications Household Survey. many regulators have assumed that the “correct” call termination rate should be set at the average incremental cost of terminating the call. July 2006 (reflecting December 2005 – January 2006 survey data). however. Valletti and Houpis (2005) find that “…the proper regulation of F2M termination charges is a challenging task and the inter-relationships between the different pricing choices and the scope of regulatory intervention are not always obvious. the size of fixed and common costs in the mobile market and the size of any (un-internalised) call externalities.128 With that in mind.

it may be possible to identify a wrong one. The service that a large Internet Service Provider (ISP) provides to a large business enterprise. The rates that are being set are nonetheless greatly in excess of efficient levels. There are strong arguments that current termination rates in Europe are too high.4. for example a DSL or cable broadband service. you are using transit. for a regulator to determine a “perfect” rate. We discuss this at length in section 5. The main conclusions that we draw are: • The post-2003 European regulatory practice is clearly superior. who was compelled to roll a stone up a hill and watch it roll down again for all eternity. raises similar concerns about the Sisyphean task129 of attempting to compute a correct termination rate: “The idea of an ‘optimal price’ is a chimera: there are as many optimal prices as there are different specifications of the underlying economic model. The transit customer pays the provider. in terms of welfare.1.2. although not necessarily by that name: If you subscribe to an Internet access service. The transit provider (the ISP) carries traffic for the transit customer. to the point where they strongly deter people from placing calls to mobile phones. The name derives from the Greek legend of King Sisyphus. 129 A Sisyphean task is unending and unrewarding. With the consultancy fees available there is no likelihood of economists running out of ideas or agreeing with each other. if not impossible. Regulation of termination charges is a Sisyphean task. is on a larger scale but is not fundamentally different from your individual transit service. With the cross-subsidies and economic rents at stake the interested parties can be expected to continue to devise new models and to argue their corner. or to a smaller or less well connected ISP. A transit service provider connects you.Final Report: The Future of IP Interconnection 69 Littlechild. 130 Littlechild (2006). to the previous practice of allowing mobile operators to set these rates unilaterally. Most readers are familiar with transit services. not the other way around. but It is difficult. to the Internet for a fee. the transit customer. . but the transit customer is not under any obligation to carry traffic for the transit provider. who is not only a prominent economist but also an experienced regulator. Transit is an inherently asymmetric model.3 IP interconnection models The majority of connections between two IP-based networks today take place according to one of two basic interconnection models: transit and peering.”130 While it is impossible to say what the right charge is. • • 3.

FCC. local telecom provisioning. and possibly also in terms of robustness and reliability. as part of the transit arrangement. with a single ISP at its root. peering has often been done on a bill-and-keep basis. where one provider (the transit provider) agrees to carry traffic to third parties on behalf of another provider or an end user (the customer). The single ISP at the root of the tree would be a potential competitive bottleneck. such as Service Level Agreements. and also a potential single point of failure. Peering thus offers a provider access only to a single provider’s customers. and to and from every destination on the Internet.org/fg/fg4/ISP_Interconnection. … Transit is an agreement where an ISP agrees to carry traffic on behalf of another ISP or end user. … is typically used where both parties perceive a 131 Report of the NRIC V Interoperability Focus Group. Transit. Peering is usually a bilateral business and technical arrangement.S. usually provides access at a predictable price to the entire Internet. … Historically. and where each party supports part of the cost of the interconnect. With peering. page 7. there is an alternative way for ISPs to interconnect. The Network Reliability and Interoperability Council (NRIC). without cash payments. then the only possible structure for a fully interconnected Internet would be a logical tree structure. an industry advisory council to the U. “Service Provider Interconnection for Internet Protocol Best Effort Service”. described131 peering and transit in this way: Peering is an agreement between ISPs to carry traffic for each other and for their respective customers. Peering is a substantially symmetric form of network interconnection. In most cases. but not for third parties. Such a tree structure would raise serious concerns in terms of competition policy. … Transit is usually a bilateral business and technical arrangement.70 Final Report: The Future of IP Interconnection If transit were the only way for ISPs to interconnect.doc. . ISPs exchange traffic for their respective customers (and for customers of their respective customers).nric. Happily. by contrast. the transit provider carries traffic to and from its other customers. and Network Operations Center (NOC) support. and from one another’s customers (and thus from their customers’ customers). installation support. the ISP often also provides ancillary services. Peering where there is no explicit exchange of money between parties. available at http://www. Peering does not include the obligation to carry traffic to third parties. where two providers agree to accept traffic from one another. In a transit agreement.

133 3. The bandwidth available to the consumer is physically limited by the size of the “pipe”.136 132 Faratin et.Final Report: The Future of IP Interconnection 71 roughly equal exchange of value. but not to all – such arrangements are the exception. Peering is often. It is also worth noting that transit arrangements exist that offer access to portions of the Internet. it is a substantially different service from transit. al. and occasionally in the academic literature. say.135 To the extent that these prices reflect the incremental capacity that the ISP must deploy.132 Peering and transit are thus distinct. large corporations). 136 Transit agreements associated with web hosting may continue to reflect traffic volume or intensity.134 Transit prices are most often flat rate per unit time. the customer access line. the simpler of the two services. notably traffic imbalance. (2007) argue that changes in the nature of the Internet. and both paid peering and partial transit have had their role to play in the Internet for at least as long. This was a perfectly reasonable system.g. in a sense. typically based on a 95th percentile of traffic samples taken at intervals of. In the mid-nineties.3. A difference is that these transit prices are usually set by the ISP in response to market forces.1 Transit Transit is. Other pricing arrangements have existed at various points in time. however. 134 In reality. traffic imbalance has been a major consideration in Internet interconnection for more than ten years. this throttling based on the size of the customer access line is less and less meaningful. particularly for enterprise customers (e. provided without cost. Peering therefore is fundamentally a barter relationship. usually dependent on the maximum traffic that the user could possibly send. no provider can guarantee access to every single IP address. many ISPs charged for transit based on actual traffic characteristics. al. . not by the regulator. mutually complementary services. Note that the NRIC definition of peering includes paid peering (which differs from payment-free peering only in the level of payment. This is especially true for broadband services to individual consumers. fifteen minutes. however. this 95th percentile measurement corresponds to a “busy hour” measurement. From the ISP’s perspective. they have much in common with Capacity-Based Pricing (CBP) in the switched PSTN/PLMN world. but not always. 135 As broadband and fibre interfaces to the home take on progressively higher maximum bit rates. and that the NRIC definition of transit includes partial transit in the sense meant by Faratin et. The transit customer pays the transit provider for access to the entire Internet. not at all in the nature of the service). which reflect near-peak traffic and is closely related with the additional capacity needed to carry the traffic (which is once again roughly equivalent to CBP). 133 There is some tendency in the trade press. providers come close enough to satisfy their customers. to treat peering as if it were effectively a transit service provided without cost. are leading to new forms of interconnection including paid peering and partial transit. but customer preference for flat rate was strong – sophisticated corporate users were willing to pay a surprisingly large premium for the certainty of knowing that they would not find one time “spikes” in their bills. not the rule. In fact.

each peering arrangement requires personnel resources to maintain over time – not only operationally. well-connected Internet service providers have no significant need for a transit provider. They are sometimes referred to as Tier 1 ISPs. Each peering relationship represents an exchange of traffic between a pair of ISPs.137 They are richly connected to one another by peering. as depicted in Figure 11 below. and thus to reduce their costs. 137 Even a large and well-connected (backbone) ISP might find it useful to maintain a transit relationship as something of a safety measure. Peering might possibly also improve the reliability and the performance of their service in comparison with use of the transit connection. • There are at a minimum thousands of independent IP-based networks in the world. and for respective transit customers of their transit customers. These transit connections typically carry very little traffic. including traffic for their respective transit customers. For all of these reasons. . The transit provider might have routes to a few destinations that would otherwise be inaccessible. This implies an unbroken chain of downstream transit arrangements from a pair of peering ISPs to their respective customers. ISPs attempt to peer with those IP-based networks where peering is most advantageous. Aside from that. no ISP would attempt to peer with all other networks. All other ISPs have a transit arrangement with some transit provider. and to use transit to reach the rest.3. These ISPs can choose to use peering as an economic optimization – they peer with another ISP primarily to offload traffic from their transit connection.72 Final Report: The Future of IP Interconnection 3. It is not technically feasible to peer with all of them. and could therefore reach substantially all Internet destinations by means of transit. Instead. but also contractually. There are also infrastructure costs.2 Peering Peering serves two distinct but closely interrelated functions in the Internet: • A few large.

and may reflect volume of traffic or near-peak traffic level. Often done without payment Source: wik-Consult. A succession of smaller ISPs have transit relationships to these Tier 1 ISPs. This roughly hierarchical structure makes the Internet as a whole cost-effective and manageable. A richly interconnected mesh of Tier 1 ISPs exists at the top of this hierarchy. some technical. together with their individual computer systems and users. some economic). Given that ISPs cannot peer with every other ISP (for a range of practical reasons.Final Report: The Future of IP Interconnection 73 Figure 18: Two peers and their respective transit customers Upstream C D ISP ISP Peering Connection B ISP ISP ISP ISP E ISP ISP Transit Arrangements A Downstream ISP ISP Parties A–B B–C E–D C–D Interconnection Arrangement Transit Transit Transit Peering Typical Nature of Agreement Bilateral Bilateral Bilateral Bilateral Typical Commercial Arrangements Payment reflects capacity. These arrangements collectively imply a roughly hierarchical structure for Internet interconnection. At the lowest tier of the hierarchy (the “leaves” of the hierarchical tree) are networks that have no customers. and that there are in most countries no obligations to peer and no set prices for interconnection. based on Gao (2000). and often have a sparse network of peering connections to one another. it is natural to wonder: • • How do ISPs decide with which other ISPs they will peer? How do they charge one another? .

they are motivated to peer under even-handed terms. For this reason. A network operator has to invest in order to reach a location at which it could “meet” a prospective peering partner. depends primarily on whether the ISP expects to save money overall (for example.) 140 On the other hand. multiplied by the distance that it is carried. but that a range of marketing and practical factors come into play.141) For this reason. an ISP may have to carry traffic received from a peer further than would be the case if the same ISP received the same traffic from a transit provider. they may be providing the competitor with an advantage that will ultimately hurt them in the marketplace. If their competitor derives more benefit than they do from the interconnection. (2007). number of IP addresses served. but also reflects competitive considerations. ISPs tend to be motivated to exchange traffic by peering with those ISPs with whom they have the greatest volume of traffic to exchange. and also to accept the transaction costs associated with managing another complex relationship. this will tend to be the largest ISPs. web hosting ISPs (which tend to send more traffic than they receive) tend to be more motivated to peer than networks 138 The discussion that follows is largely based on first hand experience negotiating peering arrangements for one of the largest Internet backbones in the world. by reduced expenditure for transit arrangements). or it could be unequal because of traffic asymmetries. that does not mean that it is without cost. Where there is rough parity. This presumes that each ISP’s costs are roughly driven by the amount of traffic carried.139 Whether a firm will choose to make these infrastructure investments. Other things being equal. Peering is often not charged for. Chapter 14. traffic sent to another network travels less distance and is thus inherently less expensive to carry than traffic received from the other network. among ISPs that use shortest exit routing. Large ISPs tend to have an especially strong preference for peering exchanges where they feel that they derive at least as much benefit as the other ISP. see Marcus (1999) or refer to the NRIC V (FCC) report on ISP interoperability (op.140 (As previously noted. cit. however. 139 For an explanation of shortest exit routing (sometimes referred to as “hot potato routing”). The reason for this is that they and their peering partner both compete in the end market for the same end-users. Due to a phenomenon known as shortest-exit routing. The decision is not just about revenue maximization. including number of customers. some large ISPs look for a rough parity in bit-Kilometres in the peering arrangements. traffic symmetry (and thus the bit-Kilometre product) is generally independent of most other measures of the relative “size” of the network. often with no money changing hands.74 Final Report: The Future of IP Interconnection The short answer138 is that firms make these decisions primarily as an exercise in costoptimization. . See also Faratin et. but they are not directly related to traffic balance or comparability of costs. which typically implies additional infrastructure investments. 141 See Marcus (1999). These factors might come into the negotiation in terms of ensuring that the parties have enough traffic to exchange to ensure that the transaction costs are warranted. and total traffic carried. The bit-Kilometre product could be unequal due to the relative geographic scope of the two ISP networks. al.

or might agree to peer only subject to payment. FCC’s approach is interesting inasmuch as it is relatively unintrusive. these guidelines have been published. (2) a sufficient volume of peering traffic to make the connection worthwhile. Table 1: Coasian arrangements: Backbone peering versus U. Mobile Operators Obligation to Interconnect Applies to all carriers. Many ISPs receive far more requests to peer than they can profitably accommodate. and Cremer.S. There are similarities to the Coasian (generally Bill and Keep) reciprocal compensation arrangements among mobile operators and non-dominant fixed operators in the United States. In order to simplify the internal decision process and to reduce the risk of disputes. Generally unconstrained. Typically. and (3) mature network operation capabilities.Final Report: The Future of IP Interconnection 75 that primarily serve consumers (which tend to receive more traffic than they send). It is for reasons of traffic asymmetry (and the resulting cost asymmetry) that web hosting ISPs sometimes find themselves in peering disputes with ISPs that primarily cater to consumers (“eyeballs”) and thus receive their traffic. If the disparity in cost is large enough. the ISP that perceives its costs as being higher might decline to peer. Backbone ISPs No regulatory obligations in most countries. Some backbone ISPs adhere voluntarily to guidelines or to principles of non-discrimination. Termination charges would still be set at levels far in excess of those that maximise welfare. Constraints on fees charged Source: wik-Consult Must by law be reciprocal (equal in both directions). There is also a tendency for small ISPs to be more motivated than large ones to peer. a generally unconstrained commercial negotiation. The reasons for this are tied to network externalities. The systems achieve somewhat similar results. Farrell and Saloner (1985). Peering is a true Coasian arrangement. Bill and Keep U. but may not be the best way forward for Europe. Rey and Tirole (2000). Our assessment is that a pure Coasian framework is insufficient to address the call termination monopoly associated with the voice service. and are explained in Katz and Shapiro (1985). . but there are also noteworthy differences. All of this implies a complex decision process. The U. some have established a set of internal decision guidelines that they use in negotiations with other ISPs. as shown in Table 1 below. Occasionally. There is no standard formula for payment. Some level of regulatory intervention is required.S. but they rest on different foundations. guidelines seek to ensure some combination of (1) rough parity of the bit-Kilometre product.S.

What is sometimes lacking in the discussion of QoS is the fact that differentiated QoS has been commonplace within individual networks for perhaps ten years now. but has not been implemented to any significant degree. We return to these points in the next chapter. The technologies used (often including DiffServ and MPLS) are not particularly difficult. and have been deployable for many years. Nonetheless. there has been keen interest in the implementation of differentiated levels of Quality of Service (QoS). It has been discussed extensively in conjunction with IP-based interconnection. not technical. implementation of differentiated QoS between networks is rare or nonexistent.3 Differentiated Quality of Service (QoS) in the Internet today With the migration to Next Generation Networks. The reasons appear to be economic.76 Final Report: The Future of IP Interconnection 3.3. .

then goes on to discuss technologies in support of data transmission and of accounting for QoS.Final Report: The Future of IP Interconnection 77 4 Differentiated Quality of Service (QoS) Differentiated QoS is closely interrelated with undifferentiated interconnection. that customers do not perceive much difference most of the time. QoS is supposed to be an important sales driver for NGN (primarily due to the integration of real time services like VoIP. • Packet protocols in general. for transmission. or packets. If an email is delayed for a few seconds. It then reviews the economic background. and are consequently not willing to pay much of a premium for better-than-best-efforts QoS. The thought was that data traffic typically is not highly delay-sensitive. but it poses unique regulatory challenges. Technical and economic considerations tend to confirm what U. but there are several that are of particular interest in an IP-based data communications network: • The end-to-end delay (reflecting both the average and the variability of delay) through the network. In the event of a momentary overload of network capacity. Providers of IP-based publicly available Electronic Communication Systems (ECS) have therefore not been sufficiently motivated to strike deals with their competitors.S. QoS will not necessarily fare better in NGN than it has in the Internet until and unless these challenges are properly understood and addressed. data could be broken up in small chunks.1 Application requirements The Quality of Service (QoS) of a network has many dimensions. and possibly also due to incorporation of multimedia and streaming services like IPTV and Video on Demand [VoD]). experience already suggests: namely. and shipped over the network as quickly as capacity permitted. users will not be greatly distressed. QoS between networks has failed to deploy. This chapter explains the application requirements that drive the need for QoS. the packets could be either delayed or else dropped outright. Consequently. and The probability of loss of a packet. were originally created in support of data transmission. or even a few minutes. and the IP protocol in particular. and has been at the centre of the Network Neutrality debate in the United States. not voice or video per se. Applications that depended on reliably receiving all packets would use communication protocols (for . however. and concludes with a discussion of Network Neutrality. QoS within and between networks has been technically feasible for perhaps ten years. There are many misconceptions regarding QoS that need to be addressed. even though QoS within networks has been widely implemented. 4. however.

it was difficult or impossible to consistently achieve the necessary delay characteristics for real time bidirectional voice. October 1990.78 Final Report: The Future of IP Interconnection example the Transmission Control Protocol [TCP]) that would deal with occasional dropped packets. . the days of the ARPANET. 142 Many assume that this is a recent phenomenon. all is well. As long as the overall probability of a packet arriving too late (which depends on both the mean and the standard deviation of delay) is sufficiently low. In an era when consumer access at home was primarily limited to dial-up access at speeds of not more than 56 Kbps. and can use the buffer to “smooth out” minor variability in packet delay. especially in conjunction with transoceanic transmission via geosynchronous satellites (where round trip delay is about 270 milliseconds). Version 2 (ST-II). The receiving system can establish a buffer of audio packets. Voice over IP (VoIP) is entirely workable. delay and loss are important to bidirectional real time voice and video. The IP protocol is optimized for applications that are somewhat tolerant of loss and delay of packets. it is a part of the normal operational behaviour of these networks. For a delay buffer to be effective. however. They are important only when both sides are interested in speaking (transmitting) – in other words. Many of the IP-based systems of the eighties and the early nineties could not reliably achieve round-trip delay of 150 milliseconds. Experimental Internet Stream Protocol. there has been interest in the transmission of voice and video over the Internet. listening to the radio over streamed audio – moderate delay and loss are not a problem as long as the listener is willing to wait a second or two for the program to begin. both the mean (average) and the standard deviation (a measure of variability) of delay must be bounded. delay or loss of packets does not represent a failure of the network. in sharp distinction to traditional telephony networks. From the earliest days of the Internet and its predecessors. because neither knows that the other has already begun to speak. The motivation for the original protocol was that IP … did not provide the delay and data rate characteristics necessary to support voice applications.” C. With today’s high speed network core. rather. variants of the IP protocol family had already appeared to “… to support efficient delivery of streams of packets … in applications requiring guaranteed data rates and controlled delay characteristics. By 1990. If delay exceeds roughly 150 milliseconds. it is possible to a point to accept a bit more variability where the average delay is low. there is a substantial likelihood that both parties to the call will try to speak at once. It is important to note at the outset that delay and loss are not a major concern for all voice traffic. In an IP-based network. and vice versa. Topolcic. and with broadband speeds to home. For one way traffic – for example. Even an occasional lost packet is not a catastrophe – the CODECs (encoding devices) are often smart enough to interpolate for the missing data. but its roots go back to the late seventies. and the human ear is good at compensating.142 The impact of delay on realtime voice was well understood from studies of telephony networks.

Differentiated QoS is. not necessarily voice and video. the implementation of QoS to support voice traffic does not enable the designer to implement a significantly smaller or cheaper network – not unless he or she is willing to intentionally sacrifice performance for the non-delay-sensitive data traffic. it will also tend to have sufficient capacity to carry the delay-sensitive voice traffic at adequate QoS under normal operating conditions.and inter-network QoS has been technically feasible for perhaps ten years. nonetheless. both delay-sensitive and non-delay-sensitive. racing. With or without QoS. potentially valuable for bi-directional real-time voice and video. small relative to the total capacity available. It is often incorrectly assumed that. Intra. although some require much more depending Key Concept Box 11: Quality of Service economics and engineering Differentiated QoS poses unique regulatory challenges. Customers do not perceive much difference most of the time. board games strategy FTP low low low high high Bandwidth Requirements Bandwidth Requirements Source: wik-Consult It is also worth noting that the amount of voice bandwidth for which delay and loss must be controlled is. QoS between networks has not been widely deployed. . and are consequently not willing to pay much of a premium for better-than-best-efforts QoS. If the core of the network is properly designed to carry non-delay-sensitive traffic. A normal voice telephony circuit requires 56 Kilobits per second (Kbps). Continued in Network Neutrality on page 94. a network needs to be properly dimensioned to carry the offered load. networks need to be overengineered with massive excess capacity in order to carry voice properly. however. Banking Email Video Streaming cards. Conversely. Internet packet protocols were originally designed to transmit data. in the absence of QoS. in many cases. The presumption is that one could implement a substantially less costly network if one were to implement differentiated QoS. A compressed audio stream could require as little as 8 Kbps. Shopping.Final Report: The Future of IP Interconnection 79 Figure 19: Application requirements for stringent QoS General Applications Stringent Latency Requirements Stringent Latency Requirements high Gaming Applications VoIP Video Telephony high sports. shooter IRC Audio Streaming Channel Surfing role-playing low Browsing.

4. or (2) the end-user is unwilling to tolerate a second or two of delay each time that a new video stream is initiated (for example. as in the case of a videoconference. and the probability of packet loss. a voice channel consumes less than one percent of available bandwidth. the impact of QoS enhancement: and the technologies that can be applied. For streamed one way video. In technical terms. Note. . as long as the end-user is willing to tolerate a second or two of delay. These video data streams thus represent most or all of the capacity of a typical broadband interface – in fact. but the amount of bandwidth required is far greater. what does this imply? How can appropriate bounds be implemented? In this section. too. which is less than the speed of light in vacuum. For video. may care about delay that would not be of concern to most users.80 Final Report: The Future of IP Interconnection on the encoding used. A compressed video stream at conventional television quality might involve 2-3 Mbps of data. no special attention to delay and loss is required. it is necessary to bound the mean and standard deviation of delay. that wires and fibres do not run in a perfectly straight line. and especially players of multi-player real time interactive games. players of online games . 4. Similar considerations apply to video. First. a buffer on the receiving side can compensate for typical variability in delay. a high definition video stream might represent 8-11 Mbps.2. as for audio. This is a function of the distance that the signal must travel. Strict control of delay is required where (1) the video is real time bidirectional or multi-directional. however. we consider the performance of IP-based networks in the absence of special QoS mechanisms. not every video stream needs to have controlled packet delay and loss.1 The performance of IP-based networks in the absence of QoS management Delay has multiple components. including both VDSL and Fibre to the building/home (FTTx). For example. Compared to a run-of-the-mill consumer broadband access at a couple of Megabits per second.2 The technology of QoS management As noted in the previous section. the end-user is “channel surfing”). For most data applications. there are exceptions. they are the prime motivation for the deeper deployment of fibre into the access network. and the speed of light143 – the 143 We are speaking here of the speed of light in a fibre or wire. there is the propagation delay through each link of the network – the time that it takes for light or electricity to reach its destination.

2. this last being a function of the packet length. See Marcus (1999). Some will be much faster. See Marcus (1999). Experienced designers compensate by designing their networks with a little extra tolerance. The more interesting delays have to do with queuing – the need for one packet of data to wait for another in order to gain access to a shared facility. a typical traffic mix for the Internet in the past would reflect a coefficient of variation (a measure of variability) of about 1. 321-322.147 The exact service time is not critical for the “wave of the hand” analysis that follows. Experienced network designers plan accordingly. The queuing delays for the transmission links are the relevant ones to consider in this case – the queuing delays through the routers (based on the router’s processing time) are small enough to ignore for purposes of this discussion. If offered load exceeds capacity (load effectively greater than 100%). or between high speed IP-based networks.144 Obviously. the percent utilisation should not exceed 90% for the busiest hour. 147 This is computed using the Pollaczek-Khichine formula for an M/G/1 queuing system. while packet length distributions may vary with the customer mix. but we can nonetheless make some general statements about these delays. 146 See Marcus (1999). and (3) to a lesser degree. The queuing delay of packets of a link is a function of (1) the capacity of the link. but queuing theorists usually work with simplified mathematical models that ignore this aspect. a standard dimension. These delays can be analysed using a branch of mathematics known as queuing theory. will generally be at least 155 Mbps. See Huston and also CAIDA (2007) . however. not thousands. The same mathematics that are used to analyse wait time in a supermarket can be used to analyse the time that packets spend waiting for a transmission link or for the processing unit of an IP-based router. pp. the large circuits inside the core of a modern IP-based network. one would reasonably expect the service time through each link in the network to average about 175 microseconds for a 155 Mbps link at 90% utilization with a coefficient of variation of packet length of 1.146 Under these assumptions. performance will be very bad. (2) the utilisation (percent busy) of the link.145 Third. there are substantial differences from one network or another. but otherwise there is nothing to done about propagation delay. The total service time of the network will typically reflect dozens of links. 148 There are some suggestions that this number is declining over time.Final Report: The Future of IP Interconnection 81 network designer should avoid needlessly circuitous routing of the traffic. the variability of the time to service the packet. an IP-based network with fast links can perform quite adequately with busy hour load in the 80-90% range.2. Second. First.148 The average delay for the network as a whole is the sum of the average 144 The detailed arrival pattern of packets can also play a role. Each IP packet travels over a network of these queues. 145 Contrary to popular belief. The unusually “bursty” nature of Internet traffic is such that the measured performance of IP-based networks can be somewhat worse than the most basic theoretical models would suggest.

The point is. to prefer some packets over others. even in the absence of differentiated QoS. on the variability of the wait time.82 Final Report: The Future of IP Interconnection delays of the individual queues. rather.2 The impact of management of QoS The network designer has a limited number of tools available to manage QoS. The normal performance of the Internet is good enough most of the time. as noted in Section 4. This is not to say that QoS is irrelevant for VoIP. Fibre-based networks experience very little packet loss due to transmission errors. To a make a long story short. the mean is not the key issue here. This explains why services such as SIPgate and Vonage function quite adequately most of the time. But the designer can choose: • To prioritise some packets above the normal best-efforts level. Differentiated QoS reduces the risk of disruption of real time voice or video due to momentary spikes in load.3. nearly all packets will arrive in time as long as no link in the path is saturated (load 100% or greater). packets are occasionally lost because buffers in routers overflow. that endusers who are willing to use mobile phones with less-than-perfect voice quality will probably find best efforts VoIP service to be quite a reasonable substitute for fixed network voice service at some appropriate price point. assuming that network designers have done their job properly. The speed of the circuits is what it is – the designer cannot make them faster for some packets than for others. In the event that packets must be dropped.2. 4. Similar considerations apply to packet loss. • 149 Per Jackson’s Theorem. .149 however. The relevant question is. and also on the fixed propagation delay (which could represent tens of milliseconds where the distances are large).2. and corrects for many other potential problems. the point at which bidirectional real time VoIP is problematic? The answer to this question depends on the number of hops (links) through the network. rather. It is often impossible for a normal end-user to distinguish between the quality of a fixed network voice call and the quality of VoIP over best-efforts IP. this will happen only rarely. what fraction of packets require more than 150 milliseconds round trip. and possibly to prioritise others below that level. except in the case where a router or link is saturated. In general.

4. or peer-to-peer file transfers – to a lower-than-nominal priority can be advantageous for analogous reasons. networks need to be over-engineered with massive excess capacity in order to carry voice properly. the packets that are disfavoured will have higher delay than in the nominal non-prioritised case. Thus. while video is likely to play a large role in bandwidth demands going forward. moreover. If the video streams represent a large fraction of the total traffic. this is likely to work well. in the absence of QoS.150 It is worth noting at this point what managed and differentiated QoS is not about. they would derive only limited benefit from prioritisation. Moreover. and that QoS will therefore prove to be essential in order to maintain the price/performance of the network. it is not excessive in comparison with the rate of improvement in the price/performance of routing equipment and fibre optics (with DWDM). . the benefits for the voice traffic (reduction of the risk of momentary overload of some circuit along a long path) are clear. We question these assumptions. Intentionally assigning traffic that does not require real time delivery – email. QoS is unnecessary (and needlessly expensive) for streamed one-way video as long as the end-user can tolerate a couple of seconds of delay at the outset. In the past few years. it is by no means clear that volumes of Internet traffic are surging more than usual due to video. The presumption is that one could implement a substantially less costly network if one were to implement differentiated QoS. but the data rate associated with that stream is fairly low in comparison with data traffic. Prioritising video presents more complex issues. If only VoIP packets are favoured. which is stil quite healthy. VoIP data is much less “bursty” than typical data traffic. the rate of growth has been slowing. substantial voice 150 These trends are discussed at some length in section 2. Conversely. by preferring one class of packets over another. for instance. not accelerating. Conversely.Final Report: The Future of IP Interconnection 83 Queuing theory provides formulae for prioritised queues. the degree to which non-voice traffic is likely to be slowed by prioritised voice traffic is not likely to be a concern. First. Internet traffic has tended to roughly double from one year to the next for most years since the founding of the Internet. Many authors assume that video is about to create an enormous surge in Internet traffic. Second.2. It is often assumed that.3. even if they will not necessarily be obvious to the user. the impact on non-prioritised traffic might be substantial. one can achieve lower average delay and also lower variability of delay. because video is voluminous. as noted in Section 4. Current traffic growth year over year is about 50-60%. Even large volumes of VoIP traffic do not represent that much data volume. VoIP represents a fairly steady stream of packets. In essence. nonetheless. If we are talking about a network that carries data.

When one or more components have failed. both delaysensitive and non-delay-sensitive. Conversely. and/or When a natural or man-made catastrophe has occurred. a network needs to be properly dimensioned to carry the offered load. Unidirectional (streaming) video does not necessarily require QoS management as long as the user is willing to tolerate a second or two of delay at the outset to fill a delay buffer. and also that performance at high speed Points of Interconnection (PoI) in the core of large networks is unlikely to benefit greatly (in the most likely scenario) from special QoS arrangements for bidirectional real time voice. for reasons noted in the previous section. With or without QoS.151 Video could potentially change the above assessment – but video would imply a demand for differentiated QoS over the interconnection only to the extent that (1) the video is real time bi. cable modem services) at the edge of the network. Where would we expect better-than-best-efforts QoS to be beneficial? Packet delay is most likely to be an issue. . rather than remaining on-net. it will also have sufficient capacity to carry the delaysensitive voice traffic at adequate QoS under normal operating conditions. 151 This statement reflects the authors’ practical experience in designing large IP-based data networks.3 Where management of QoS is likely to be beneficial The discussion in the previous section suggests that differentiated QoS is not required for all applications. For shared access facilities (e. 4. and perhaps even streaming video. Real time bidirectional voice and/or video clearly do. or (2) there is a strong demand to provide instant response to channel surfing.1. not all applications benefit from better-than-best-efforts QoS.2.g. and QoS management thus most beneficial: • • For slower circuits at the edge of the network. and (3) the video traverses the interconnection point. If such a network core is properly designed to carry non-delay-sensitive traffic. this is simply incorrect. the implementation of QoS to support voice traffic does not enable the designer to implement a significantly smaller network – not unless he or she is willing to intentionally sacrifice performance for the non-delaysensitive data traffic.or multi-directional. When one or more circuits are saturated.84 Final Report: The Future of IP Interconnection traffic. Many data applications do not. • • • As noted in section 4.

4.2. DiffServ and MPLS are trivial to implement within a network. considers the ongoing evolution of standards in regard to QoS management. in the end-user’s own network – it may not be cost-effective to send the video in real time across an interconnection between networks.2 Differentiated Services (DiffServ) In response to concerns over the complexity of RSVP and the ISA. DiffServ enables hop-by-hop traffic management. the IETF evolved a much simpler communications protocol referred to as DiffServ (for Differentiated Services).2. which tends to be voluminous. It was eventually shut down. the video will often be stored or cached closer to the user. It was a technical success but a commercial failure. The common wisdom has been that these protocols were hopelessly complex.4 Technologies used to manage QoS A number of technologies are in widespread use already for management of QoS. Where an enduser initiates video on demand. 4. 4. and are in use in many large networks today.2.2. where selected packets . For reasons noted in section 2. In fact. and then explores the implications of differentiated QoS for network management and Operational Support Systems (OSS). It was delivering services to real customers. but not every video stream will need managed QoS across an interconnection PoI.1 The Integrated Services Architecture (ISA) and RSVP In the early nineties. and MPLS/GMPLS.Final Report: The Future of IP Interconnection 85 Again. but rather because the company never found enough customers who were willing to pay much of a premium to use the RSVP-capable network. due in part to implementation complexity.3. (Non-real-time transfer from the content provider to the cache does not require managed QoS. at least one Internet Service Provider (BBN) had a working network based on RSVP. the IETF was active in evolving a series of relatively complex solutions under the rubric of the Integrated Services Architecture. RSVP never saw widespread deployment. DiffServ.) Video will certainly be a factor for managed QoS.4. linear data may continue for quite some time to be transmitted using non-IP-based cable television and PON capabilities.4.1. QoS will obviously not be needed at the PoI unless data is actually sent across the PoI. The IETF developed a number of mechanisms for QoS management. This is particularly relevant to video data. as exemplified by the RSVP protocol. not due to technical problems. This section considers each of these technologies. notably including RSVP.

4.4. it implies surprising complexity. . they would need to verify that each actually delivered services within its own network at the level of quality that it had committed to the other. It is up to each router. billing and accounting The implications for Operational Support Systems (OSS) in support of differentiated QoS tend to be overlooked in most discussions.2. and to deal with discrepancies between the measurements of the parties. It is a natural complement to DiffServ. To the extent that this implies the need to account for QoS-capable traffic. then.86 Final Report: The Future of IP Interconnection can be marked as having application requirements other than best efforts. there would be the need to reconcile statistics. it can provide adequate overall assurance at a statistical level. Technologists tend to focus more on the problem of getting the bits to move as they are supposed to move. a pair of network operators would need to agree on how much QoS-relevant traffic each delivered to the other. each would need some tools to deter fraudulent use or “gaming” of the system. Still. 4.152 Finally. DiffServ provides a more limited service than RSVP in the sense that it assures performance only on a hop-by-hop basis.3 MPLS/GMPLS Multi-Protocol Label Switching (MPLS) is a technique and protocol that can be used to implement differentiated QoS. to do what it can to implement the desired transmission characteristics (or to decline to do so). and less on the problem of how to ensure that someone pays for those movements. with Multi-Protocol Label Switching (MPLS) being perhaps the most common today.3. The first is trivial. Second.2. It has generally been assumed that a network operator would be willing to provide better-than-best-efforts quality only to the extent that either the end user or another network operator were willing to pay them a premium to do so. 4. and when – otherwise. Capturing first-order statistics on traffic sent between the parties is straightforward. First. there is 152 At a minimum. Measuring traffic across a link would seem to be straightforward. Finally.4. MPLS and GMPLS are discussed at length in Section 2.4 Implications for OSS. rather than end-to-end. the second and third are difficult. Various techniques can then be used to ensure hop-by-hop performance. and distinguishing among different marked classes of traffic is no harder. Even here. some prior agreements would be needed as regards what is being measured. they would need to be able to perform spot checks so as to deter systematic violations.1.

these networks are likely to be direct competitors for the same end users – network B is not about to let network A place sensors in its network. and neither will want to rely on measurements provided by their respective end users. and if so who might that trusted third party be? The challenges in verifying that the quality of service that was actually delivered was consistent with the quality committed are much more profound. It might possibly be far simpler to bill. Network A thus needs reliable performance statistics about network B’s network. even though both are measuring (different ends of) the same link using substantially similar tools. the one that is sampling at more frequent intervals will tend to see an apparently more lumpy distribution). and for implementing measurement regimes.” If the providers do not agree. Reconciling data would be challenging.Final Report: The Future of IP Interconnection 87 the risk that network A has a slightly different view of the traffic delivered on the link from A to B than does network B.2. not for the use of the network.5 The ongoing evolution of QoS-relevant standards The core IP-based technologies that are being deployed for QoS have changed only marginally over the past ten years. And sampling intervals need to be mutually agreed. The MIT white paper was developed with substantial industry participation. Each network is likely to be skittish about providing internal performance data to one another. 2006. the service provider will not necessarily be the network provider. At the same time. Take one or three.1. there are challenges – in an IPbased NGN. There is an old Dutch proverb: “Never go to sea with two compasses.edu/resources/papers/Interprovider%20QoS%20MIT_CFP_WP_9_14_06. It seeks to develop a standard framework for standardized QoS classes. In this case. 4. too.pdf. . available at http://cfp. standard deviation) are likely to reach different conclusions due to the perverse effects of the Central Limit Theorem (if two sensors sample the same distribution. otherwise any measures of variability (quantiles. and for end to end 153 White Paper draft 1.4.mit. but rather for the services that benefit from differentiated QoS. Here. November 17. whose statistics should govern? Is there scope here for a trusted intermediary. Particularly noteworthy in this regard is the Inter-provider Quality of Service white paper153 developed by the Communication Futures Program (CFP) at the Massachusetts Institute of Technology (MIT). Interest continues to focus on DiffServ and on MPLS. network A needs to ensure that network B delivered the committed performance. There has been substantially more positive movement in regard to the development of practical QoS metrics for use between providers.

These are measured using IPPM probes – one way measurement packets sent at irregular (Poisson-distributed) times. one suitable for bi-directional real time VoIP. page 23. It is not the whole story – the most daunting challenges to deployment of inter-provider QoS continue to be economic rather than technical. the economic transaction costs of reaching agreement may be reduced. the MIT paper potentially represents a significant advance in achieving inter-provider QoS.g. one way variability of delay. drawing heavily on the IETF IPPM metrics that have been stable for many years. By providing a standard framework for discussion and negotiations between providers. By providing standard service classes and measurement methodologies. As a useful simplifying assumption. or optionally a slightly more expansive group of networks including both core and access IP-based networks. Inter-provider Quality of Service. the other oriented toward conventional best-efforts service. Still. The MIT work assumes multiple customers seeking services among a group of at least two interconnected networks. they assume only two classes of service. and one way packet loss. It focuses on end-to-end metrics. The White Paper defines suitable end-to-end values for one way delay. draft 1. .88 Final Report: The Future of IP Interconnection measurement. These metrics are then apportioned to the multiple networks involved in delivering the service (e. the MIT white paper represents important progress. with an average of five per second.1. Figure 20: MIT reference model for inter-provider QoS Source: MIT CFP. Provider A and Provider B in Figure 20 above).

3 The economics of differentiated QoS This section introduces the basic economic factors that are relevant to differentiated QoS.1. In some cases. the new service starts with a zero market share. many industries experience network externalities. .2 Network externalities As explained in section 3. and may be largely unrelated to underlying costs. In consequence.Final Report: The Future of IP Interconnection 89 4. This is true of telephone service.3. By definition. we take it for granted that we may be offered first class and second class tickets. My telephone is worth more if there are a great many people whom I can call. but they manifest themselves somewhat differently in conjunction with differentiated QoS. rather. My Internet connection is worth more when there are a great many people to whom I can send an email. Deregulation of the airline industry. and finally. network externalities. incentives for interconnection.154 When we buy a ticket for a train or an airplane. this kind of price discrimination tends to enhance consumer welfare. and also of the Internet. When an airline offers cheaper tickets to passengers who are willing to stay overnight on Saturday. then.3. with a higher price for the former. It usually seeks to replace a service that has a substantial market share. it has nothing to do with their costs. the 154 See Hotelling (1929). and a great many websites to which I can connect. First. In the absence of market power. it reflects the greater willingness to pay (lower elasticity of demand) of business travellers. Many of these are familiar themes from Chapter 3. 4. price discrimination may be linked solely to the willingness of the customer to pay. 4. Getting a new service launched in a sector dominated by network externalities can be challenging.1. A service may be most useful when a great many people use it (and not just because of economies of scale). and by offering different qualities at different prices to different groups of customers. is generally acknowledged as having resulted in lower prices for consumers.1. then.1 Service differentiation It has long been recognized that providers of goods or services could potentially achieve some pricing power and profitability by distinguishing their goods and services. but are in most cases unwilling to stay overnight outside of the Monday to Friday time frame. transaction costs. we explore service differentiation in general. Business travellers are willing and able to pay more. and emergence of price discrimination.

firms are in general motivated to be fully interoperable. the value of deployment might be significant to those networks that implement it later. monitoring tools. The service has some value within a network. Thus. If it were available to only two or three networks. It might have great value if it were available to every destination on the Internet. and coordination in general would need to be put in place.155 Agreements. one firm has a large enough market share.3 Transaction costs Extending differentiated QoS to each additional network implies transaction costs.156 4. then that firm will prefer less-than-perfect interoperability. If.3. it is hard to get the process started. These concerns are not unique to differentiated QoS. Thus.90 Final Report: The Future of IP Interconnection externalities associated with the pre-existing service keep pulling the sector back to the old.3. then it is of limited value. The adoption of Internet Version 6 (IPv6. because perfect interoperability would dilute the firm’s ability to exploit the market power that its large market share confers. 4. a new version of the Internet Protocol with a greatly expanded address range) and of DNSSEC (a security enhancement to the Domain Name System) have arguably been only glacial deployment due to similar network effects / transaction cost problems.4 Incentives for interconnection A recurrent theme in the discussion of IP interconnection is whether network operators will be motivated to interconnect (on reasonable terms) in the absence of a regulatory obligation. but the initial benefit to the first two or three networks to deploy it is minimal.157 155 The concept of transaction costs was introduced in Section 3. A number of Internet capabilities are faced with similar economic challenges. Full interoperability implies the largest possible target market. and thus tends to be to the benefit of all market players. There is a body of economic literature (most notably.3.1. 156 See Marcus (2004a). in a market characterized by strong network effects. Differentiated QoS between and among networks is subject to these network effects. both in absolute terms and relative to its next largest competitors. and away from the new. however. Katz and Shapiro (1985)) that demonstrates that. For a given network. . 157 See See Katz/Shapiro (1985) and Farrell/Saloner (1985). and it would be hard to get it to completion once it had been launched. these costs might be roughly linear in the number of networks with which that network maintains QoS-capable arrangements.

it would have deployed. Cremer/Rey/Tirole (2000) applies the Katz and Shapiro (1985) analysis to the question of the incentives that large Internet backbones have to peer with one another.2 has shown. keeping superior QoS as a competitive advantage for their own affiliated offerings? Are there countervailing forces that limit the ability of providers with market power in their ability to execute anticompetitive strategies? 4. why has it been so slow to achieve deployment between and among networks? Differentiated QoS is not a new idea.158 As Section 4. The technology has been basically mature for a decade. how? In the absence of regulation. For reasons noted in Section 4. the idea has technical merit – not as much. 159 Assuming. but economists subsequently realised that the same analysis applied with equal force to network interconnection. the basic answer is obvious: Had the benefits of deployment clearly exceeded the costs. that network operators with market power do not choose to degrade it. From an economic perspective. perhaps.Final Report: The Future of IP Interconnection 91 Katz and Shapiro (1985) was formulated in terms of standards compliance. would these operators refuse to interconnect with their most capable competitors? Might they block access to certain destinations? Might they intentionally and perhaps selectively degrade the quality of the interconnection? Might they relegate competitors to best-efforts service. but merit nonetheless. These considerations manifest themselves in a number of concerns as regards the quality of interconnection that IP-based network operators who have market power in one form or another might prefer. . there are indeed questions as to whether the perceived benefits are too low.3.5 Slow deployment between service providers to date Given that the technology of differentiated QoS is not particularly challenging. as some proponents might claim. one might infer that either that the perceived costs are too high.159 In addition. and given its widespread use within networks.2 of this paper. Thus. or that the perceived benefits too low. a series of challenges related to network externalities and to transaction costs have inhibited deployment. of course. or perhaps both. and if so. Might they be motivated to exploit their market power in anticompetitive ways. 158 We speak from first hand knowledge. and the idea was extensively discussed by large backbone Internet service providers in the US many years ago. end-users are likely to perceive QoS as being adequate for VoIP most of the time. Even in the absence of managed QoS.

4 The Network Neutrality debate in the United States and elsewhere A new form of market power is possible in the IP-based world.3.2. Users are unlikely to perceive a difference.161 160 See Rohlfs (2003). even in the absence of any special QoS management technology.3. 4. It is difficult to get past the initial adoption hump in order to achieve critical mass.3 Transaction costs For reasons noted in Section 4. QoS can make a greater difference elsewhere – for example. 4. Katz/Shapiro (1985) and Farrell/ Saloner (1985). CDs were successful because Matsushita and Phillips had commercial interests in both CD players and studios.g. for reasons explained in Section 4. as a function of the network effects associated with control of access to large numbers of customers.4. in shared media (e. 4. or to outages). and were thus motivated to ensure that both players and content were available.. VCRs were initially purchased for time-shifting of television programs. it will tend to be motivated to implement less-than-perfect interconnection and/or interoperability in order to exploit its advantage. each QoS-capable interconnection agreement requires non-trivial management attention to maintain. .1.2 Benefits are often not visible to the end-user The performance of traffic between large ISPs will tend to be excellent most of the time and under most circumstances.5. to poor planning. only when enough consumers had purchased VCRs did a video rental business emerge.3.3 above. Some services never get past the initial adoption hump. where a single firm controls a sufficiently large share of the user base. getting a new service launched in a sector dominated by network externalities can be challenging. a body of economic theory says that. for example.3. network operators have been reluctant to make the necessary investments. In the absence of sufficiently strong demonstrated customer demand. to traffic “spikes”.3. and are therefore unlikely to be willing to pay a substantial premium. The economist Jeffrey Rohlfs160 has explained that different services historically got past the initial adoption hump in different ways.5.1 “Bandwagon Effects” and network externalities As previously noted. 161 See Cremer/ Rey/ Tirole (2000).5. As noted in Section 4. cable television broadband systems) at the edge of the network.92 Final Report: The Future of IP Interconnection 4. or where links are saturated (due.

Final Report: The Future of IP Interconnection

93

The concern that has most often been expressed is that a network operator might either block or degrade access to disfavoured content in order to favour its own content or affiliated content. For example, a network operator might strike a deal with one online bookseller, and impede access to competing booksellers. In one concrete instance, a small dominant wired operator blocked access to the Vonage VoIP service, apparently in order to hamper a competitor to its own conventional voice services.162 To date, there are few concrete indications of exploitation of this form of market power. The emergence of the Network Neutrality debate in the United States and (with somewhat less vehemence) in Europe reflects legitimate concerns over possible abuse of market power in connection with the Internet in the absence of regulation. It is no accident that this debate, which had been simmering for years, only reached a boil when three interrelated things happened163: (1) the FCC eliminated obligations of nondiscrimination toward content for providers of broadband access to the Internet;164 (2) the wholesale ADSL market segment in the U.S. collapsed165; and (3) the U.S. telecommunications industry underwent significant consolidation, where two of the largest Internet backbones were acquired by presumably dominant local incumbents.166 Violations of Network Neutrality could be viewed primarily as an anticompetitive manifestation of market power. The objectionable forms often represent a form of economic foreclosure, where a firm that possesses market power in one market segment attempts to project that market power into an otherwise competitive market segment upstream or downstream. At the same time, widespread violations of Network Neutrality could conceivably impact free speech and media pluralism in the longer term.

162 Consent degree in the matter of Madison River Communications, LLC and affiliated companies, available at http://fjallfoss.fcc.gov/edocs_public/attachmatch/DA-05-543A2.pdf. 163 See Marcus (2005a) 164 FCC, In the Matters of Appropriate Framework for Broadband Access to the Internet over Wireline Facilities; Universal Service Obligations of Broadband Providers …, adopted August 5, released September 23. 165 Less than 4% of those Americans who use DSL broadband Internet access get their access from a wholesale provider (a CLEC), according to FCC data, and this percentage is declining year over year (see FCC (2007)). A substantial fraction of Americans get their broadband access from cable, but there is no wholesale offer. 166 SBC acquired AT&T, taking on the latter’s name in the process; Verizon acquired MCI.

94

Final Report: The Future of IP Interconnection

Key Concept Box 12: Network Neutrality

Network Neutrality is a many-faceted problem. A common concern has been that a network operator might either block or degrade access to disfavoured content in order to favour its own content or affiliated content. For now, we do not see a need for comprehensive new regulation to address Network Neutrality challenges. European underlying markets are either competitive or well regulated, and the European regulatory framework provides adequate tools to address problems should they emerge. Modest improvements, such as those proposed by the Commission on 13 November 2007, seem to be appropriate as a means of ensuring that consumers can exercise informed choice. Continued in Net Neutrality: US vs. Europe on page 96.

4.4.1 Quality differentiation, social welfare, innovation, and market power
Many so-called “violations” of Network Neutrality should not in reality be viewed as problematic. In the sometimes confused debate that is taking place in the United States, there is a tendency to view offers of different grades of QoS to end-users as if they were somehow insidious, when in fact many are probably either neutral or perhaps consumer-welfare-enhancing. As noted in Section 4.4.1, price differentiation (in the absence of market power) can enhance consumer welfare. As an example, one widely cited paper on violations of Network Neutrality167 suggested that charging extra for a static IP address was anticompetitive. In fact, a static IP address facilitates the ability of the end-user to host IP-based services (such as a web site), which may increase the cost to the network. The static IP address also presumably correlates with willingness to pay. It is rational, and not necessarily anticompetitive, to charge a premium. The prospect of service differentiation should not be viewed as a problem per se. It is the risk of economic foreclosure that is a potential problem.

167 See Wu (2005).

Final Report: The Future of IP Interconnection

95

4.4.2 Vertical integration and other vertical relationships between network operators and applications and content providers
Vertical integration between network service providers and web-based content providers has not automatically conferred advantages in the past. When America Online (AOL) purchased Time Warner, there were widespread concerns that the combination of AOL’s Internet content with a major cable television operator (that was also a major broadband ISP168) would harm consumers. The U.S. FTC and FCC went so far as to impose merger conditions entailing wholesale access to cable broadband capabilities, and interoperability with AOL’s Instant Messenger Internet chat service. In reality, however, the merged firm has found it very difficult to exploit synergies between cable broadband Internet access and Internet content. The acquisition of Excite by @Home provides a similar lesson. @Home, a firm which provided support to the Internet Service Provision operations of several of the largest cable broadband service providers in the U.S., purchased Excite, a then-large search engine and portal. There are no indications that the cable companies attempted to make it difficult for their customers to access competing search engines and portals (such as Yahoo), nor is there any suggestion that Excite subsequently tried to limit access to @Home’s broadband users. Again, the parties fail to exploit any synergies. These results are not inconsistent with the Katz-Shapiro theory, if one assumes that the firms involved were below the necessary market share threshold. The prediction is that they would seek to maximize profitability by serving the whole market.

4.4.3 Implications for European policymakers
In the U.S. context, these concerns are real; moreover, they cannot easily be fixed through regulation. The problems are too complex. It is too difficult for regulators to distinguish between those forms of service differentiation that enhance consumer welfare versus those that harm it. Moreover, any legislative or regulatory fix is likely to impact innovation. In Europe, by contrast, the underlying markets are much more competitive; moreover, the regulatory system in Europe is likely to ensure that they remain competitive. Even in relatively concentrated European broadband markets, most consumers can choose among multiple broadband Internet Service Providers (many of them service-based rather than facilities-based). For the most part, the network neutrality debate has not emerged in Europe, and it is unlikely to emerge in the same form in which it has in the United States.

168 In the United States, a majority of broadband Internet access was and is provided by cable television companies.

the Network Neutrality debate has a very different character in Europe than it has in the United States.. we consider the risk of network providers intentionally degrading service. either selectively to disfavoured applications and content or else to their best-efforts IP services overall. If this potential problem were to emerge with full force in Europe. Whether this should be done in advance.3. Continued in VoIP peering on page 97. and the European regulatory framework provides a much richer range of tools with which to address competitive bottlenecks. Europe Europe enjoys a more robustly competitive market for broadband access than does the U. As a result. NRAs and NCAs probably have adequate tools to deal with it.96 Final Report: The Future of IP Interconnection Key Concept Box 13: Net Neutrality: US vs. In Section 5. it might be appropriate to impose non-discrimination obligations on incumbents with SMP. is a complex judgment call. or whether the regulator should instead wait to see if the harm really appears.S. At the regulatory level. The regulator could also require the operator to make QoS information publicly available pursuant to Articles 20 and 22 of the Universal Service Directive. . since intentional degradation would appear to represent a form of anticompetitive foreclosure or tying. The alternative to a regulatory ex ante solution would be to instead approach this kind of intentional degradation as an ex post competition law matter.

which is already widespread. then it goes on to cover NGN. Voice over IP (VoIP). and of the NGN are not really distinct phenomena – they are all aspects of technological and market convergence. which appears to be poised to achieve widespread deployment. • VoIP peering is likely to exist as something physically and economically distinct from IP peering. VoIP peering becomes increasingly attractive due to positive externalities (more VoIP users to connect to) and the efficiency gain (priceperformance) that derives from avoiding multiple conversions from VoIP to circuitswitched and back. • VoIP service providers can interconnect at lower cost and with greater efficiency by using VoIP peering with one another rather than PSTN interconnection. raising economic and regulatory issues. of IPTV. of IPTV. and how to deal with cost modelling when costs increase due to parallel operation before they decline due to improved long term price/performance. .Final Report: The Future of IP Interconnection 97 5 NGNs. • VoIP peering must deal with identifying the service provider responsible for a number. It includes a discussion of the technical structure of NGN. There are many uncertainties as to how today’s networks will in fact evolve. and IP interconnection This chapter discusses applications that may necessitate new approaches to IP interconnection. even if their ISPs are not directly interconnected. • Interconnection at the IP level could also facilitate new services such as video conferences and presence services. which has become complicated due to number portability. Key Concept Box 14: VoIP peering The evolution of VoIP. what to do about changing numbers of points of interconnection. cont’d on page 100. and of the NGN are not really distinct phenomena – they are all aspects of technological and market convergence. • As the number of VoIP users grows. First. it covers VoIP. Continued in VoIP Peering Issues. especially in connection with IMS/TISPAN (which potentially poses the risk of new bottlenecks at the Application Layer). and closes with a discussion of key migration and transition issues: how long to maintain SMP remedies. but it seems safe to predict that services will continue to migrate to an IP base. It is important to note that the evolution of VoIP. The chapter includes a discussion of different migration paths to NGN.

see no need to invest in traditional telephony database approaches. European Communications. Now. VoIP peering can respond to the need to determine the provider of VoIP service for a particular telephone number. for instance. 2006. but one of the most significant is easily overlooked. Moreover. the number itself encoded which service provider must be used.1.co. Instead. Casema. using VoIP.eurocomms. as most analysts now expect. and exchanging traffic with other VoIP providers purely at the IP level. 169 See. For IP-based providers. and as the corresponding likelihood increases that a given VoIP call will be received by a VoIP service or phone. “UPC Netherlands.uk/1319.1 Interconnection of VoIP In interconnecting islands of VoIP. this can be done more cheaply and more reliably using IP-based tools rather than traditional PSTN databases. PSTN-to-PSTN interconnection is mature. large blocks of telephone numbers tended to be assigned to service providers according to a publicly understood allocation scheme – when a company wished to complete a call to that number. however. There are a number of motivations for VoIP peering.1 Rationale for VoIP peering The primary driver for interconnecting VoIP services at the IP level is the gain in efficiency (and thus in price-performance) that derives from avoiding multiple conversions from VoIP to circuit-switched and back.169 At the same time. and PSTN-to-VoIP interconnection is reasonably well understood. indeed. it is increasingly common for techniques to be used beyond those of conventional IP peering. 5. These firms. the transition to that world is not well understood. January 26. at: http://www. the association is no longer reliable – each individual number must be somehow looked up a in database. they will build their new voice businesses directly on their broadband Internet access capabilities. and must deal with the further complexities of number portability. Historically. VoIP peering can take place even if the VoIP providers are themselves using ISPs that are not directly interconnected. . much less work has been done to date on the likely evolution of these arrangements at both a technical and an economic level in the event that IP becomes. the traditional switched PSTN/PLMN world will be with us for many years to come. VoIP peering becomes more attractive as the number of VoIP users grows. MultiKabel. VoIP Peering has been of particular interest to the cable television industry in North America and in Europe.98 Final Report: The Future of IP Interconnection 5. the basis for the preponderance of telephony. some of which are new entrants to telephony. Essent and CaiW jointly select XConnect to exchange all VoIP traffic”. as a result of number portability.

it may be possible to use the traditional PSTN mechanisms for these look-ups. Finally. ENUM is a mature standards-based mechanism. When VoIP interconnects with the switched PSTN/PLMN in the traditional way. the operators have found it necessary to construct increasingly sophisticated relationships between telephone numbers and the end-point of services. ENUM was not specifically designed as an aid to VoIP peering. In interconnecting different VoIP services.323.3 ENUM and VoIP peering One look-up capability that has been of particular interest in this regard is based on (carrier) ENUM. the phone number must be looked up (much as a DNS name is looked up and mapped to an IP address) in order to find the call routing information with which the telephone number is associated. Where VoIP operators interoperate with one another. with the advent of number portability. to map a telephone number to a ranked list of Internet services. These conversions are sometimes accomplished by a Session Border Controller. conversion among SIP. but it happens to represent a relatively straightforward approach that drew on pre-existing stable technical standards. Over time. In effect. the historical mapping breaks down altogether. 171 For example. large blocks of telephone numbers were associated with individual carriers.Final Report: The Future of IP Interconnection 99 This is not to say that no conversions are required. and a more customized look-up mechanism is required. 5.1. drawing on the technology of the Domain Name System (DNS). and that needed very little new technical 170 A CODEC is a coder-decoder device that encodes. it is possible that there will be a need for conversion from one CODEC170 to another. This approach was simple and logical in an era when telephone numbers mapped one to one to physical lines.1. it is more natural to use other look-up mechanisms. 5.2 Number portability and VoIP peering Historically. . These assignments were effectively part of the national numbering plan. which might also serve as a security frontier (firewall) and/or a measurement and control point to the VoIP service provider’s network. however. in this instance. Many different CODEC standards are in use today. and MGCP. At that point. IP-based look-up mechanisms can offer not only improvements in operational efficiency for the interconnection of voice services. or from one VoIP signalling protocol standard171 to another. the relationship between a phone number and the physical location at which service is provided begins to look more like a database query problem. in connection with (for example) freephone services and premium services. H. but potentially also interoperability for video and presence services. analog voice into a digital signal.

The original ENUM standards contemplated a publicly available service where anyone could look up Internet services associated with a particular phone number. but not all connected individuals will wish to be listed. ENUM is said to be considerably more efficient than traditional telephony look-up mechanisms. after much Key Concept Box 16: Carrier ENUM and Public ENUM Public ENUM is publicly and globally visible. Continued in Mobile Interconnection on page 105. The Internet Engineering Task Force (IETF) created the ENUM standard and defined a zone in the public DNS space where public ENUM will reside. Individual privacy rights must be respected. however. there has been interest in private carrier ENUM systems. not the “public” ENUM envisioned by the original IETF standards. thought. VoIP service providers). Carrier ENUM in its nature must list all connected individuals. which raise fewer concerns as regards user privacy. in principle. Carrier ENUM. . Continued in Carrier ENUM and Public ENUM below. however. Public ENUM is designed to be publicly and globally visible. In the case of carrier ENUM. cont’d • ENUM is a mature standards-based mechanism.172 Key Concept Box 15: VoIP Peering Issues. • ENUM could be particularly useful as an aid to VoIP peering. it would be a private ENUM. they declined to define a space for carrier ENUM. 172 This use of ENUM is referred to as carrier ENUM. the information need not be made available to the general public. but will be available only to service providers. visible only to the carriers. based on ENUM technology but accessible only to carriers (in this context. Individual privacy rights must be respected. drawing on the technology of the Domain Name System (DNS). but individuals can choose not to be listed. visible only to service providers. For this reason. to map a telephone number to a ranked list of Internet services by means of which the user can be reached.100 Final Report: The Future of IP Interconnection development. uses the same technology but lists all users. Carrier ENUM and public ENUM use the same technology. Carrier ENUM is likely to continue to evolve for this purpose. but it will evolve in a somewhat different direction from that of the original “public” ENUM. In addition. but they will not necessarily be the same implementation. it is of interest only to other carriers.173 It is likely that the International Telecommunications Union (ITU-T) will define a DNS zone for carrier ENUM.

1. or bundled with the overall price of connectivity.2 reflects on specific incentives of fixed and mobile operators to offer IP-based interconnection to their intrinsic voice services.4.1. the ISP’s own application services are typically either charged to the end-user of the service at retail.Final Report: The Future of IP Interconnection 101 5. primarily in the form of third party VoIP.4.1 What the literature implies about VoIP peering The economic analysis of IP-based interconnection in Section 3. If a ISP is paid.1 considers the issue from a economic perspective.4.3 discusses what we know from our interviews and literature review as regards the likely evolution.1. These issues have important public policy implications. If so. however.1. what economic arrangements are likely to prevail for VoIP peering? What charging and accounting arrangements are likely to prevail for IP-based services. Presumably.1. as distinct from data services. this area is beginning to see active research as a result of the network neutrality debate. These are the same mechanisms that are in play when independent third parties offer application services – the payment flows from the end-user to the application service provider.1. relatively little economic analysis had been undertaken of interconnection challenges at the upper levels of the network architecture. we do not see service-based wholesale payments today either because ISP network operators lack sufficient market 173 See http://www1. Section 5. Section 5. also reflect the QoS requested.4. They also play a significant role in our conclusions and recommendations in Sections 6 and 7.4. Charges at wholesale typically reflect just the capacity provided and/or the volume of traffic transmitted. In the past.4 evaluates the degree to which competitive alternatives. independent of the underlying network? What market power concerns might these charging arrangements raise? What are the implications for regulators? Section 5. it is for the traffic it carries.ietf. respectively. and perhaps the QoS associated with that traffic – both of which are directly related to the ISP’s costs. which we touch on in Section 5. paid for by advertising. but generally not the fact that a particular user was invoking a voice or video service. Charges could.3 of this report addresses IP traffic exchange. might drive the evolution of these arrangements. 5.org/.1. primarily in the context of peering and transit. in principle.4.4 VoIP peering and market power Will VoIP peering exist as something physically and economically distinct from IP peering? There is reason today to think that it will. Today. To the extent that an ISP also provides application services.5. . we do not seem to see separate wholesale payments between Internet Service Providers (ISPs) for IP-based services that ride on top of their respective networks. Section 5.

as long as only a single operator is in a position to terminate a call to a single E. inherent ability of the Internet to use alternate paths.3. . all or which are either irrelevant to conventional voice service. closer to the application.176 This argument is clearly incorrect. Fixed and mobile incumbents who benefit from current termination arrangements will attempt to preserve them for as long as possible.3.1. The paper provides several additional rationales. Consequently. and especially to IMS.164 telephone number. the party to use a different network. and especially mobile operators.164 telephone number.4. or else appear to be factually incorrect. Under CPP/CPNP arrangements. for the same reason that circuit-level alternate routing is irrelevant in a conventional network. to attempt to prolong their termination revenue. 176 See Reynolds et al (2007). section 7. for reasons of technical efficiency. 175 See Cullen International/Devoteam (2003) and Elixmann et al (2007). the termination monopoly will persist. or because the existing pricing mechanisms are sufficient. or both. Horrocks (2005) suggested that interconnection payments should. One recent paper argues that the Under CPP/CPNP arrangements.2 Incentives of mobile and fixed operators At one level. 5. and (2) the termination monopoly will persist. might create new forms of market power in the upper layers of the network. is most often implemented without payment between the parties — in the absence of separate payments at the application level.174 Wholesale payment arrangements at the application layer going forward have been difficult to predict. The inherent routability of IP traffic is equally irrelevant. it is easy to make an economic prediction.102 Final Report: The Future of IP Interconnection power to successfully demand them. A number of papers175 have suggested that the migration to NGN. 174 This discussion of Network Neutrality in section 4. as termination monopoly will become less long as only a single operator is in relevant as voice moves to an IP-based a position to terminate a call to a world due to (1) the ability of the initiating single E. evolve to a system where network layer payments for traffic carried should be quite independent of application layer wholesale payments. The caller’s ability to initiate the call from a different network is irrelevant to the monopoly on termination.4 is relevant here. Internet peering. one might expect conventional switched PSTN/PLMN operators.3. a migration to IP-based interconnection might threaten or end this cash flow. as we saw in Section 3.

uses the same GSM technology as Europe. but also reflects network externalities..179 We expand on this point later in this section.182 and • 177 This section is closely related to the discussion of the “right” termination rate in section 3.S. or both – any or all of which argue that call termination rates are in excess in costs.4.-style system. either as a means of gaining market share or of disadvantaging rivals. 181 Littlechild (2006) provides a regression of these same Merrill-Lynch data. Singapore and Hong Kong.S.S. worse in others. in order to reflect externality benefits of the party receiving the call. the CDMA IS-95 that is used in the U.S. .177 A number of specific observations support this view: • Many European mobile operators offer much lower prices for on-net calls than for off-net mobile-to-mobile (M2M) calls. Canada. there is good reason to believe that they perceive their termination revenues to be greatly in excess of their true average incremental cost. Alternatively.181 Given that the mobile industry in Europe is generally considered to be competitive. with minutes of use four times higher in the U. According to Merrill-Lynch data on the wireless industry. but the difference cannot explain these differences in price. Note that the Merill-Lynch data appear not to count on-net mobile-to-mobile calls in CPNP systems (because they do not generate revenue.” By RPP. he found a difference of about $0.180 For most Member States. service based revenue per minute is more than twice as high as that of the U. so the difference presumably is mostly a reflection of termination rates. Technology cannot explain this difference – to a significant degree. Singapore and Canada. one would expect that retail prices would otherwise be set to competitive levels. 179 See Armstrong and Wright (2007).S. For the remainder. or of strong network externality benefits to the recipient of a call. See section 3.] cents per minute compared to CPP) and significantly increases average usage (by about 143 minutes per month).S.2. The apparently large on-net/off-net price difference is strongly suggestive of a difference in perceived costs. might do better than GSM in some circumstances.3. depending for example on terrain. because the mobile operator does not pay a termination fee to itself). normalized service-based revenue. and may thus somewhat understate MoU and correspondingly overstate service-based revenue per MoU. Littlechild found: “RPP significantly reduces average revenue per minute (by about 12 [U. service-based revenue per minute of use (i.2. they offer flat rate pricing for calls to the fixed network and for on-net calls to mobile operators.e. 178 Some plans are effectively “buckets of minutes” plans.S.4. Harbord and Pagozzi (2007).178 Ignoring for the moment network externalities. This on-net/offnet price difference is not solely a difference of cost.S. With externalities in mind. mobile operators will see some competitive benefit in price differentiation between calls to on-net and off-net mobile. In a more detailed comparison of the UK and US. 182 About 45% of the U.S.. Armstrong and Wright (2007) find that the welfare-maximising termination rate would be less than marginal cost (and much less than current European rates). mobile market is GSM. There is no evidence that RPP lowers mobile penetration rate.Final Report: The Future of IP Interconnection 103 In the case of mobile operators. and Littlechild (2006).14 per minute. incorporating other potential explanatory variables. but rather about how mobile operators perceive the profitability of the termination rate. 180 This is consistent with Littlechild (2006). See Figure 17 in this report.3. however. which includes revenues from termination) is typically between 50% and 400% higher in European Member States than it is in the U. Littlechild refers to countries with a U. this strongly implies that they themselves see the termination payments that they must make to other mobile operators as being greatly in excess of their own marginal costs to terminate a call. the U. including the U. the discussion in this section is not about social welfare. which means that they function as a banded flat rate. but not for off-net M2M calls to other mobile operators.

according to Armstrong and Wright (2007). Nor can teledensity explain this difference. Again. the welfare maximising termination rate should in any case be lower than marginal cost in order to reflect externality benefits to the recipient of the call. . Retail rates often converge (for plans at the high end of usage) to an incremental retail price of € 0.04 per minute. (2007). • A recent WIK study found real average incremental cost per minute for an Australian mobile operator to be in the range of € 0. 183 See Brinkmann et al. had traffic volumes been higher.186 One would not expect costs in Cyprus to be significantly lower than in other EU Member States. The US incremental price is [so low as to make] only a negligible contribution to the variable cost of outgoing calls. [This in turn suggests] that the variable cost of making or terminating mobile calls is considerably lower than CPP regulators commonly assume.0225.03 to € 0.” • • These observations support the notion that mobile operators in Europe probably view the termination fees that they receive as being well in excess of their true marginal cost.02 per minute. European rates seek to reflect average incremental cost. We understand that the rate is about € 0.185 Cyprus.114.012 per minute. Littlechild (2006) finds the marginal retail price of T-Mobile in the U.S. “The level of the incremental price for the higher volume plans may provide some indication of the incremental cost of termination. which is well under € 0. 184 See EU-Commission (2007) 185 Today. 186 See ERG (2007b). As Littlechild puts it. nor labour costs. Presumably their mobile operators are not providing service at a deficit.183 In comparison with average European mobile termination rates of € 0.104 Final Report: The Future of IP Interconnection the frequency bands are not that much different.01 per minute. in the largest “buckets” to be $0. has a mobile termination rate of some € 0. no operator is likely to sell service at less than its marginal cost. Again. incremental cost might have been a mere € 0.02 or less per minute for mobile minutes in countries with low or zero termination rates. this suggests that the variable cost itself cannot be very high. a Member State of the EU.184 these costs are quite low.02 per minute.

for prices to be so much higher suggests the presence of economic distortions. They argue that the expected on-net price is adjusted downward to reflect externalities.. the simplistic interpretation (still ignoring network externalities) would be that the correct termination rate should have been negative (i.1 pence 187 See also Harbord and Pagnozzi (2007).Final Report: The Future of IP Interconnection 105 Key Concept Box 17: Mobile Interconnection Key conclusions include: • • • • The post-2003 European regulatory practice is clearly superior. however. 5. Given that the termination rate at the time was 5. Expanding on the discussion of on-net / off-net pricing differences. The rates that are being set are nonetheless greatly in excess of efficient levels.03 to € 0. The difference in the two prices therefore is exactly the amount by which the termination fee exceeds the marginal cost of terminating the call. but not for off-net calls to other mobile operators.02. for a difference of 7. had traffic volumes been higher.1 pence.1 pence)!187 Armstrong and Wright (2007) go on to provide a more comprehensive analysis that considers network externalities. the expected price of off-net calls is the cost of origination plus the termination fee (ignoring network externalities). Service-based revenue per minute of use is typically between 50% and 400% higher in European Member States than it is in the U. There is no obvious reason why costs should be more than twice as high as in the US on average. • • Continued in Mobile IP Interconnectionon page 108.04 per minute. to the previous practice of allowing mobile operators to set these rates unilaterally. and are probably well in excess of real usage-based marginal cost. while on-net M2M calls had an average retail price of 4. but It is difficult. if not impossible. for a regulator to determine a “perfect” rate. . Singapore and Canada. A number of observations support the observation that termination rates are too high: Many European mobile operators offer favourable or flat rate pricing for calls to the fixed network and for on-net calls to its mobile network. in terms of welfare. This strongly suggests that there is a significant difference between the incremental cost that the mobile operator perceives for calls on its own network in comparison to the payments that it makes to complete calls to other mobile operators.S. incremental cost might have been € 0. while the expected off-net price is adjusted upward (for reasons that they describe as anticompetitive).3 pence. A recent WIK study (Australia) found real average incremental cost per minute to be approximately € 0. Armstrong and Wright demonstrate (consistent with others) that the expected price for on-net calls is equal to the cost of origination plus the cost of termination. In conjunction with the 7. They also note (quoting Ofcom) that off-net M2M calls in the UK in 2005 had an average retail price of 11.e.9 pence.9 pence minus 7.2 pence.

193 One possible explanation for this seemingly counter-intuitive result is an economic phenomenon known as double marginalisation. and low or zero monthly fees (the “waterbed effect”). Note. Also. for instance. 192 EBITDA for European mobile operators does not seem to be conspicuously higher than for their U. This is so because minutes of use per month is ten times higher in the U.3. and given back to customers in the form of handset subsidies. is about 2. this leads to the same conclusion: the termination rate that has been set is greatly in excess of the efficient rate. introductory offers. . incidentally. In the case of Germany. even though national spectrum authorities do not necessarily seek them. we are not necessarily claiming that the mobile operators are making windfall profits – that is a more complicated question. and a question that is well beyond our remit in the current study. which depress use. and thus also suffer under current European arrangements. 193 See Marcus (2004). even though service based revenue per minute is four times higher in Germany than in the U.S. thus depressing ARPU and total revenue.191 It is not necessarily the case that shareholders in European mobile operators simply pocket the benefits of greater profitability – it is entirely possible that European investors in mobile telephony receive a proportionately similar slice192 of a much smaller pie than their U. it is quite clear that mobile operators could increase ARPU by lowering their prices. this implies that the externality value to the recipient is large. than in Germany. they still harm consumer welfare by distorting pricing structures. At the same time. This is generally consistent with Genakos and Valletti (2007).S. that there is no consensus as to what constitutes an optimal termination rate.106 Final Report: The Future of IP Interconnection difference in price between on-net and off-net calls.S. that even to the extent that profits are returned to customers though subsidies. presumably as a result of the difference in retail price per minute. Returning to the claim that European termination rates are too high. as we did in section 3. both to consumers at retail and to 188 See.S. they noted that the analysis was difficult because so many operators are on both the fixed and the mobile markets. Bomsel et al (2003). as a notable example. 191 ARPU in the U. Note that handset revenues and subsidies do not constitute service based revenue. 190 The high prices bid for European spectrum might themselves constitute an additional level of multiple marginalisation.5 times higher than in Germany.4. termination fees in Europe are lower today than they were at the time. 189 See Littlechild (2006). A number of financial analysts have observed the same over the years. Some of the profits associated with termination revenues are presumably being competed away.189 The higher revenues may also motivate mobile operators to incur higher costs in their auction bids for spectrum. or too much customer churn.190 There is another reason why high termination fees do not necessarily imply windfall profits. If one follows the reasoning of Armstrong and Wright (and many others) that the welfare-maximizing termination rate reflects the cost minus the network externality value to the call recipient. We would caution.S. however. counterparts. They found that termination fees were not all being returned to consumers. however. which interacts with the analysis in complicated ways. The high termination fees lead to high retail prices. where prices that are too high are charged in two vertically related markets (in this case.188 Some may be burned off as inefficiency in the form of too-rapid handset replacement.2. and that there was moreover a net transfer away for users of fixed telephony. counterparts.

we have not determined that this is the case. but they help to enforce high retail prices for all market players.Final Report: The Future of IP Interconnection 107 other network operators through termination fees). By default. it is likely to be the case that all or nearly all consumers will have broadband Internet access over their fixed service at home (if they have fixed service at all). 196 See GSM Association (2006). 195 This manifests somewhat differently for fixed-to-mobile versus mobile-to-mobile. and to prevent (for example) third party VoIP providers from connecting to a mobile operator’s standard voice service. Note. one might expect third party VoIP providers to be excluded. As networks evolve. we merely observe that it is a possible explanation. Our point in this section is a narrower one – that European mobile operators will tend to view their termination fees as exceeding their costs. We infer that SIP servers will also be part of the IPX infrastructure.3 Technical evolution of VoIP interconnection arrangements Based on the foregoing section. but we assume that access to the GRX will be limited. The economic literature refers to this as a tacitly collusory outcome. it represents direct revenue. and will therefore be motivated to preserve them. both consumers and suppliers suffer. and mobile operators have apparently implemented and are using the GRXes. however. There has been some discussion that fixed operators might also be allowed to use the GRX system. For fixed-to-mobile. The GSM Association has developed standards for an interconnection capability for mobile networks (PLMN operators) known as the GPRS Roaming Exchange (GRX)196.1. GSM-A specifications make clear that the addresses will not be 194 See Carter/Wright (2000). Is it possible for mobile or fixed operators to enforce separate interconnection arrangements for voice (as distinct from IP traffic in general) in such a world? In the case of the mobile operators. Again. an answer is emerging. The IP exchange clearly has the ability to treat VoIP voice interconnection as something distinct from IP interconnection in general. potentially usable worldwide. this is required. Carter/Wright (1999) and Carter/Wright (2003). In a world with number portability. GRX is in the process of being enhanced to become an IP eXchange (IPX).195 5. See Laffont/Rey/Tirole (1998a). and all are likely to seek to differentiate themselves. it is clear that mobile operators (and possibly also fixed) will be motivated to maintain separate interconnection arrangements. The IP addresses used in the GRX/IPX are public IP addresses. that we are not suggesting that all mobile operators have fully aligned interests as regards mobile termination fees – small operators may have different preferences than large. .. and that they will be able to access IP-based services from their mobile handsets. The IPX will include carrier ENUM services that can be used to look up the service provider who serves a particular mobile number. For mobile-to-mobile. incidentally.4.194 Under double marginalisation. the termination fees tend to be roughly neutral to net revenue.

It enables cascading billing. We are not necessarily suggesting that the decision to hide GRX/IPX IP addresses from public view is an inappropriate or malicious design decision.” The annex on IPX requirements makes the same point.197 In other words. Generally Internet routers shouldn’t know how to route to the IP addresses advertised to [GRX] networks. • The IPX clearly has the ability to treat VoIP voice interconnection as something distinct from IP interconnection in general. • The IP addresses used in the GRX/IPX are potentially usable worldwide. • The GRX/IPX employs a “hubbed” architecture which can afford the advantage of mitigating transaction cost problems in getting QoS-capable services launched among multiple providers. not only for wholesale pricing. the implications for service interconnection as mobile voice resources migrate to VoIP are clear. The GRX is in the process of being enhanced to become an IP eXchange (IPX). and to restrict access to the IPX. These arrangements have the positive property that they tend to reduce economic transaction costs by making it possible for a VoIP service provider to conclude a single contract with the IPX hub. a mobile carrier industry group. rather than individual pairwise contracts with every other VoIP service provider present at the exchange. will be accessible only to mobile network operators and possibly also to other firms that accept MNO pricing and quality arrangements. We expect that key voice infrastructure. 197 Ibid.108 Final Report: The Future of IP Interconnection reachable from the public Internet. A chain of billing can be established from the application and through each IPX hub to the end user. The IPX represents a billing model as much as a technical model. GSM Association specifications make clear that the addresses will not be reachable from the public Internet. including carrier ENUM and SIP servers. however. the hubbing also raises the concern that IPX commercial arrangements might serve in effect to enforce collusion. however. In oither words [GRX] networks shall be totally separated from public Internet. but perhaps also for retail pricing. Key Concept Box 18: Mobile IP Interconnection The GSM Association. . mobile operators will not need to actively block third party access to their own mobile capabilities – key functionality that third parties would need to access these capabilities will be maintained on networks that are completely blocked from access by organisations other than those that participate in the GRX/IPX (presumably members of the GSM-A and perhaps others by special arrangement). Section 4. At the same time. as shown in Figure 21. [these] networks shall remain invisible and inaccessible to the public Internet. however.2 is quite clear: “For security reasons. has developed the model of a GPRS Roaming Exchange (GRX) for an interconnection capability among mobile networks exclusively. Continued in VoIP Interconnection on page 112. these arrangements raise the concern that the IPX might serve to effectively “lock in” existing mobile wholesale charging models at the very moment when changes in network architecture would otherwise be calling them into question.

Both fixed and mobile SMP operators will presumably continue to be obliged to offer PSTN-based interconnection to their conventional voice services. even where the same end-users have IP data connectivity through the same fixed operator. until and unless alternative IP-based interconnection is available in a form that meets the needs of competitors. 198 See Ofcom (2005).Final Report: The Future of IP Interconnection 109 Figure 21: End-to-end SLAs and charging with the GSM-A IPX Source: GSM-A website One knowledgeable interviewee expressed concerns that fixed incumbent operators (and possibly also competitors) would likely also base their own voice services on private (inaccessible) IP resources as they evolve their voice infrastructure from PSTN to VoIP. This would have similar implications for the fixed network as for the mobile – third party VoIP service providers would not be able to access the standard voice services of fixed operators using IP interconnect.198 The implication would appear to be that mobile operators. or (2) they are under a regulatory obligation to offer it. except perhaps by special arrangement. will not offer IP-based VoIP interconnection to their standard voice services to third parties unless either (1) they believe that migrating to IP-based interconnection will have no negative effect on termination revenues. . and probably also fixed operators.

it would presumably put pressure on the traditional operator to evolve their interconnection arrangements to better accord with the new technical and market realities. it is important to remember that there are many different forms of VoIP. In framing this issue. Skype is a familiar example of a self-provisioned service. Arbitrage often serves to enhance consumer welfare in this way. . Clearly.110 Final Report: The Future of IP Interconnection 5. but permits a somewhat more integrated offering since the service provider offers both the VoIP service and the broadband access service. The column on the right is of particular interest here. inasmuch as it is independent of the network service provider. and a broadband Internet connection paid for by the customer in order to provide VoIP for a fee. Commercial VoIP services such as SIPgate (sometimes termed Voice over Internet) typically employ equipment at the customer’s premises. since a self-provisioned VoIP is by definition not affiliated with the access provider. thus bypassing the traditional operator.4 Competitive alternatives The discussion in the previous section relates to the ability of third party VoIP service providers to access the voice services of traditional mobile operators and fixed incumbents as the traditional switched PSTN/PLMN operators evolve their networks to IP-based NGNs. Figure 22 categorizes VoIP services according to two major criteria: 1) whether or not the service is self-provisioned by the end-user and 2) whether or not the VoIP service provider is affiliated with the access provider.1.4. Were this to become widespread. Solutions provided by a fixed network operator or a cable television provider are conceptually similar. This yields three forms of VoIP rather than four. To the extent that third party VoIP is effective. it would tend to generate arbitrage opportunities that would help to correct for market or regulatory failures. This is an important question from a public policy perspective. the end-users could potentially achieve an equivalent result in a different way by choosing third party VoIP services themselves.

manipulate pricing structures so as to make third party VoIP less attractive. Third. usually with substantial subsidies when purchased as part of a new mobile subscription or pre-paid service. This appears to happen more often in the U. but it happens. Other than the termination fees that they must pay. and they price accordingly at retail. a great many handsets are provided by the mobile operator. First. In Europe. Second.Final Report: The Future of IP Interconnection 111 Figure 22: Different Forms of VoIP Implementation Network Service Provider Selfprovisioned Third Party Service Provider N/A Skype Commercially Provisioned Voice over Broadband Voice over Internet SIPgate Vonage Source: wik-Consult For the fixed network in Europe. 199 Also Vonage in the U.S. For the mobile network. . 200 The degree to which this is true varies significantly from on Member State to another. the analysis is far more complex. these third party VoIP operators have very little cost. Third party VoIP service providers like SIPgate and Skype199 provide a reasonably good substitute for fixed voice services.200 The mobile operator could disable or undermine the effectiveness of VoIP services in the handsets provided to customers. Third party VoIP services face significantly greater challenges in the mobile environment than in the fixed. the ubiquity and mobility offered by mobile operators represents an enormous and legitimate advantage. if they chose to. mobile operators have considerably more control over their handsets than do fixed counterparts. the prospects are reasonably good (but by no means certain) that consumers will continue to be able to access third party offerings.S. than in Europe. the mobile operators could.

To the extent that third party VoIP is effective. and it may be appropriate for a conference that the end-user attends. consumers on the fixed network are likely to continue to be able to use third party VoIP service providers. The mobile operators have a range of tools to try to slow or block take-up of third party VoIP if they choose to do so. IP interconnection will clearly be the preferred solution for IP-based voice as well as data. or a coffee shop where he or she has a subscription to a WiFi service.112 Final Report: The Future of IP Interconnection If mobile operators are not enthusiastic about VoIP capabilities in the handsets that they provide. . TeliaSonera. Continued in Internet vs. From a technical and an economic perspective. but it does not appear to represent a mass-market challenge to the mobile operators. taking advantage of the mobile operator’s ubiquity and support for full mobility. how should it be used? Many handsets support WiFi. Traditional fixed and mobile operators will not wish to evolve their voice services to IP interconnection until and unless they see that change as being neutral or positive to their interconnection revenues. The alternative is to use the VoIP software over the mobile operator’s own IP service.g. This. it would tend to generate arbitrage opportunities to bypass the traditional operator that would help to correct for market or regulatory failures. the user could download and install VoIP software (e. this seemingly simple installation can nonetheless require a few hours of work on the part of a fairly sophisticated user. Key Concept Box 19: VoIP Interconnection As mobile operators and fixed incumbents evolve their networks to IP-based NGNs. the evolution to IP interconnection for voice could easily be thwarted by economic considerations. compared to 32% placed at home and 24% at work. See Sami Ala-Luukko. “IPX – A Key Enabler for IP Communication Subsystem”. For many of these mobile phones. Today. VoIP over WiFi is a fine solution for the end-user’s home.. but not truly mobile. It is less useful for an end-user who is constantly changing locations. Once the VoIP software has been downloaded. and an additional 11% while travelling. September 2006. however. leaves the end-user entirely at the mercy of the mobile operator’s pricing 201 One source estimates that 14% of mobile calls in Western Europe are placed while in an automobile. and not useful at all for a user who is walking down the street or travelling in a train or automobile. NGN on page 117.201 VoIP over WiFi is a solution that is nomadic. Third party VoIP services could face significantly greater challenges in the mobile environment than in the fixed. however. Skype) himself or herself. Arbitrage often serves to enhance consumer welfare in this way. or unless they are required to do so. It is surely possible. manufacturers may not be motivated to invest in provide robust capabilities with good user interfaces.

but you can see that there are powerful reasons for the cellular network providers to stay in this regime for a while. or else have to explain a massive loss of revenue to their shareholders. In a recent paper. However. 204 See. Jon Crowcroft observed: “I have a cell phone. It uses around 14kbps to carry voice. • 202 See Wu (2007). . for instance. This is much more the case for mobile operators than for fixed.4. and also leaves the end-user vulnerable to possible impediments to access.1. even though underlying technical and market drivers will have changed.. 203 See Crowcroft (2007).202 The possibility that mobile operators might intentionally price so as to discourage the use of VoIP cannot be ignored. The key lesson here is that legacy service providers resist the pressure to become merely bit pipes. I 'dial-up' over GSM as it is cheaper). I can run Skype or any vanilla VoIP system on this fairly easily. there are more cell phones than Internet Hosts (2. Indeed. EDGE and 3G as it happens). The risk that the mobile operator might block or degrade access to third party VoIP is perhaps limited in Europe due to the effectiveness of regulation.Final Report: The Future of IP Interconnection 113 plans for IP packets. the volume and time tariff of the data service is set such that a normal pattern of voice calls made over it would cost more than the GSM service.”203 For all of the reasons noted in this section. The 3G service runs at around 384kbps in the UK. but it appears to routinely be the case in the United States. and provides a global service which is extremely pervasive and affordable.204 The analysis in this section of the report suggests that voice services may evolve in a different direction from IP services in general – indeed.I do not know the architecture of the backhaul network once the wireless segment of a route is terminated. we see signs that this is already taking place. Key observations are: • Traditional switched PSTN/PLMN operators are likely to have both the incentive and the ability to maintain voice termination arrangements on the current model (or on a model more akin to the current one) for an extended period of time. and seems to have pretty low latency -. 5. mobile operators are likely to have far more ability than fixed operators to resist changing their business models to conform to new economic and technical realities as the network evolves to an IP-based NGN.5 billion active mobile phone numbers in the world at the time of writing). usually when I read my e-mail via my phone. My cell phone provides data (GPRS. [emphasis added] This is fairly surreal (in fact. but it seems to support pretty close to zero loss. Marcus (2006b) and Marcus (2007).5 Implications going forward A number of previous studies have tended to focus on pure IP connectivity as networks evolve to IP-based NGNs.

it raises questions as to the proper role of public policy in fostering a migration to IP-based interconnection. existing regulatory policy may have the unintended consequence of preventing an otherwise cost/benefits-enhancing infrastructure upgrade. the evolution of voice interconnection and data interconnection are already diverging. These specifications primarily address the separation among the service provision. however. The work was initiated by the Joint Rapporteur Group. as long as market power issues have been addressed. IP interconnection is clearly the preferred solution for IP-based voice as well as data. Second. They will not choose to evolve their voice services to IP interconnection until and unless they see that change as being neutral or positive to their interconnection revenues (unless they are required to do so). First.2 Technical characteristics of NGN The ITU has specified NGN in two basic Recommendations • • Recommendation Y. “General principles and general reference model for next generation networks”.2001 (12/2004) “General overview of NGN”. In other words. the principle of technological neutrality would suggest that the regulator should permit network operators to interconnect as they see fit. • • The public policy implications are rich.114 Final Report: The Future of IP Interconnection • These networks are already providing IP interconnection for their data services. What is the proper response? We return to these questions in our recommendations and conclusions in Chapters 6 and 7.2011 (10/2004). From a technical and an economic perspective. and was subsequently promoted by the NGN Focus Group that was created in the middle of 2004 and completed its . In this case. The distortion in the pricing structure implies a loss of consumer welfare. with or without a migration to IP-based interconnection. and does not exclude the possibility of a loss of supplier welfare as well. and Recommendation Y. the evolution to IP interconnection for voice could easily be thwarted by these economic considerations. 5. the observation that mobile termination fees are too high is a problem. Normally. control and bearer transport plane and the ability to support a wide range of services and over a range of access media using standardized (open) interfaces.

Final Report: The Future of IP Interconnection 115 work in November. The vertical functional layering reflects the layered model that the Internet Engineering Task Force (IETF) developed for the TCP/IP protocol suite. for the architectural core of the Internet. Figure 23: The ITU’s layered NGN architectural model Infrastructural. 2005. Currently. This vertical layering makes it possible for different market players to participate at different layers in the value chain of service production.e. i. application. Resources Transport management functions Transport control functions . as a result of the migration of global networks to layered network protocols based on IP. the migration to NGN is accelerating and reinforcing this tendency. middleware and baseware services Services NGN service Resources Service management functions Service control functions NGN transport Transfer functional area Source: ITU (2004). These changes in the value chain were taking place long before NGNs came on the scene. however. The functional architecture of an NGN provides a vertically layered scheme where the application and associated content (“services”) are on the top and the physical transport and access (“transfer”) are on the bottom (see Figure 23). ITU standardization work on NGN is under the NGN Global Standards Initiative (NGN-GSI).

but in contrast to the Internet reflects a centralized control platform containing corresponding call agents (soft switches) for the different types of services. 5. This approach can support access by a range of devices and corresponding access networks by means of corresponding access gateways. the NGN concepts that many incumbent operators are implementing (or propose to implement). support for different grades of Quality of Service (QoS) – this model of service deployment provides no inherent solution. To the extent that transport services are needed that are not inherently present in the network – for example. or of new software into end systems. it is viable only to the extent that the underlying networks already provide the kind of connectivity that is needed. without making any special arrangements with network operators. that seeks to employ standard.1 Interconnection at various levels “Telco-style” approaches to the integration of new services into an IP-based platform are significantly different from those that are typical for the Internet today. It is the emergence of inexpensive intelligent. reusable modules for commonly used functions. often by the addition of new servers. as well as a service/application creation environment. where a service provider might achieve market entry by simply deploying SIP Servers and (for privacy reasons) Session Border Controllers and connecting them to the Internet.116 Final Report: The Future of IP Interconnection The ITU further divides the layered model of network architecture into three planes: • • • the user information transport plane. the signalling/control plane. programmable devices for the network periphery that made this approach viable. services are generally implemented at the edge of the network. .205 which enables parties other than the network operator to offer new services. based on IP and SIP protocols. For the Internet. IMS is a forward-looking standards-based platform. user access could be contingent on the current load in the IP 205 See Isenberg (1997). This is consistent with the Internet philosophy of a “dumb” transport network and an intelligent periphery. At the same time. and the management plane. such as BT’s 21CN. VoIP is an obvious example. it can also support the integration of mobile access by means of the IP Multimedia System (IMS). This platform contains an IP-based core transport platform. seek to integrate currently separate networks into a single network and service platform. By contrast. Thanks to the soft-switch concept. at the same time.2.

but also the implications for competition in instances where the NGN operator might have significant market power.Final Report: The Future of IP Interconnection 117 transport layer. and is possible due to the emergence of inexpensive intelligent. thus enabling network design and dimensioning in line with Grade of Service (GoS) and QoS parameters. NGN Internet services are generally implemented at the edge of the network. Key Concept Box 20: Internet vs. This platform contains an IPbased core transport platform. Continued in Conclusions on page 130. but in contrast to the Internet approach reflects a centralized control platform where call agents (soft switches) provide a service/application creation environment. The architecture of the NGN seeks instead to integrate networks and services that today are often separate into a single network and service platform. a pure service provider with little or no infrastructure can easily offer services at a national or even international level using the public Internet. All of this implies the need to consider not only general technological and economic factors. the integration of new services into an NGN (based. for instance.2. Drawing on TCP/IP. and thus depends on the latter’s willingness to open up the NGN to independent service providers. This vertical layering potentially enables different market players to participate at different layers in the value chain of service production. both at the transport and at the service layer. The two concepts (Internet versus NGN) imply different interconnection strategies. By contrast.2 IMS and NGN – standards and implementation The IP Multimedia System evolved as an attempt by the mobile industry to “bolt on” mobility capabilities into the NGN architecture. often by the addition of new servers or software. The two are now reasonably well integrated. This section considers first technical standards. For example. . The functional architecture of an NGN provides a vertically layered scheme where the application and associated content (“services”) are on the top and the physical transport and access (“transfer”) are on the bottom. programmable devices. This is consistent with the Internet philosophy of a “dumb” transport network and an intelligent periphery. then implementation of IMS. on IMS) requires strong coordination with the NGN operator. much as in legacy telecom networks. 5. as is the case with Skype or SIPgate.

IMS was developed primarily in order to provide multimedia services over third generation mobile networks.2. finalized in March 2002. NGN technical standards were developed primarily by the ITU and by ETSI. but they have largely converged. however. Like NGN. p. 3GPP subsequently developed improved versions of IMS in releases 6 (wireless access) and 7 (fixed access). Thus. as shown in Figure 24 below. Later on. IMS is based on end to end IP services controlled by the SIP protocol. The IP Multimedia Subsystem (IMS). Current ITU recommendations for NGN are based on IMS (NGN-IMS) as incorporated into ETSI TISPAN. This similarity of structure facilitated the incorporation of IMS into the NGN standards.118 Final Report: The Future of IP Interconnection 5. the IMS standards are in the process of being incorporated into ITU and ETSI NGN standards. 6. originated with the 3GPP.1 IMS Technical Standards The NGN and IMS standards originated independently. the IMS is a layered architecture.g. UMTS in Europe. but with some differences (mainly in the QoS provision scheme). applications and content. the European Telecommunications Standards Institute (ETSI) incorporated IMS into the NGN specifications developed by ETSI Telecoms & Internet converged Services & Protocols for Advanced Networks (TISPAN). IMS first appeared in 3GPP release 5 specifications. a standards body focused on mobile networks. but extends them in order to enable open access to value-added services. Figure 24: Layered view of the IMS Model Source: Kinder (2005). but only for mobile access. It thus adds session control functions . e. IMS provides the functions of a SIP-based soft-switch.2.

Final Report: The Future of IP Interconnection 119 so as to enable the seamless use of multimedia services from different access technologies. The Network Attachment Subsystem (NASS) which provides the Network Attachment Control Functions (NACF). as shown in Figure 25 below: • • The IMS core. both control and transport. by means of the Interconnection Border Control Function (I-BCF) in the control plane and the Interconnection Border Gateway Function (I-BGF) in the transport plane. including authentication and authorization of the user. and to enable users to enjoy seamless roaming (including fixed-mobile roaming). The internal architectural structure of IMS comprises three major elements. fixed and mobile. as shown in Figure 26. • Figure 25: Functional elements of IMS Source: Knight (2006) In ETSI TISPAN. thus promoting fixed mobile convergence. . TISPAN defines these elements in order to enable NGN IMS operators to apply control mechanisms at entry to their respective networks. and The Resource and Admission Control Subsystem (RACS) which provides the Resource Attachment Control Functions (RACF). including resource management and admission control based on the user’s profile and the resources currently available. the interconnection scheme between the IMS 3GPP and the TISPAN IMS is done in both planes.

one is a Guaranteed QoS (and thus absolute). Note that the 3GPP IMS operates (according to the standards) only under IP version 6. delay or packet loss) could be an issue in the interconnection between the ETSI NGN IMS and 3GPP IMS. the other is a Relative QoS. Should IPv4-IPv6 translation be necessary. Provision of QoS (defined in terms of bandwidth. This means that 3GPP provides a relative QoS. . it is the job of the I-BGF. while NGN IMS supports both versions 6 and 4. 3GPP IMS defines separate QoS traffic classes that are handled according to operator requirements. The ETSI TISPAN IMS has two approaches for QoS control. The user might not receive the expected QoS.120 Final Report: The Future of IP Interconnection Figure 26: Fixed and mobile interoperability under ETSI TISPAN Source: Moro and Fernandez (2005). Conflicts might arise when a user in the NGN world subscribed to a service with Guaranteed QoS connects to a user/server/service in the 3GPP IMS world with relative QoS.

2. The strategy of the fixed operators was to evolve towards the NGN with the softswitch at the heart of the network. They need to open new business lines because the traditional business model is nearly exhausted as mobile penetration approaches 100% in first world countries. This migration is largely driven by loss of market share to mobile operators and VoIP services. Note that there are three independent paths depending on the starting point of the network operator.2. around 2004 they realized that the evolution in the core network of the mobile operator networks. . This led them to charter the ETSI TISPAN working group to adapt the IMS concept to the network architecture of the NGN operators. however.Final Report: The Future of IP Interconnection 121 5. The second evolutionary path is driven by mobile operators. Figure 27: Evolutionary paths to IMS NGN Source: wik-Consult The first path is associated with both PSTN operators (many of which already provide IP data services). pushed by 3GPP in Releases 6 and 7. would nonetheless leave them in second place in the telecommunications market. They therefore seek to evolve the access networks using enhanced technologies as High Speed Packet Data Access HSPDA. and the core network with a unified IMS control platform for service delivery.2 Evolution to IMS/NGN Figure 27 shows likely evolutionary paths for different types of networks.

2 Prospects for third party services over NGN Relative to the services offered.1 Implications of standardized NGN equipment Relative to the sale or lease of hardware and software for use in the public network.3 NGN/IMS technology and interconnection at the service level and the IP level The introduction of NGN. However.122 Final Report: The Future of IP Interconnection What is more or less certain at this point is that the evolution of both fixed and mobile operators will pass through the IMS. Standard interfaces will tend to increase the ease with which products of one supplier can replace those of another. The implications for consumer welfare are complex. there will be a smooth evolution. 5. 5. . introduces a number of new standards-based compatibility interfaces into the public network. and generally enhancing consumer benefits. It is not so clear that external access to IMS capabilities is the only acceptable way to achieve that goal. thus enhancing global competition.3. inasmuch as their infrastructure is already inherently IP-based. Finally. The good news is that these standard interfaces open a great many new avenues for competition in regard to services offered to end-users. This change may tend to reduce the pricing power of equipment suppliers. advancing the Single Market.3. this will not be done all at once. It is clear that regulators would hope to see competition at the application layer of the NGN. the migration is generally positive for consumer welfare. rather. and less revolutionary. This path is more evolutionary. and therefore are likely to limit the use of these capabilities. the third evolutionary path comes from the world of Internet service providers. 5. but will correspondingly reduce costs to service providers and should thus increase consumer surplus (assuming that markets are competitive). but also less standardized. This conclusion follows from the same economic arguments (Katz/Shapiro (1985)) that we have made repeatedly in this report. The bad news is that service providers who have market power (either in the classic sense or as a function of network externalities) presumably will not want to open their networks to competition at the service level. and of IMS in particular. the story is more complex. The underlying concept is different – it is less centralized.

For reasons noted in Section 4. the user’s inherent ability to access a third party VoIP provider (such as SIPgate or Vonage) offers a capability that is functionally equivalent to CPS.2.g. . best efforts IP service will deliver these services at quite acceptable performance unless IP-based network operators are allowed to intentionally (and perhaps selectively) degrade the performance of their best efforts services (either by intentionally crippling them or equivalently by failing to obtain necessary capacity upgrades).Final Report: The Future of IP Interconnection 123 In an IMS system. this is not the only way to provide services in an NGN. Directive 2002/22/EC. In an IP-based and undegraded world. Network operators apparently anticipate that they will have an advantage to the extent that they can provide better QoS to their own applications. 5. European regulators and competition authorities appear to have adequate tools to address intentional degradation were it to emerge. Assuming that incumbents cannot intentionally degrade successfully. will be available over best efforts IP. the migration to an all-IP broadband-based platform may render Carrier Selection (CS) and Carrier Pre-Selection (CPS) remedies206 irrelevant. We return to this point in the discussion of Network Neutrality in Section 4. If operators choose to. they can prevent third parties from offering such services.3. Article 19. then it would seem to be premature (and possibly not proportionate) to consider a regulatory intervention to open interfaces such as IMS to third parties. However. it is not yet clear whether the incumbent’s integrated IMS/NGN capabilities represent an unacceptably high barrier to entry to independent IP-based providers of applications and services over the network. 206 Access and Interconnection Directive. A range of services.1. this advantage may not be decisive – under most circumstances.4.3 Implications for Carrier Selection and Carrier Pre-Selection As a related matter. The end-users will presumably still have best efforts IP-based Internet access. but with greater flexibility. or could perhaps extract supracompetitive rents from the application service providers. including VoIP and IPTV. At this point. It does not exactly correspond to call-bycall carrier selection (from a single handset). through LLU). IMS can definitely serve as a gatekeeper relative to applications provider by the network operator. less complexity and presumably much lower cost. but one might reasonably expect technical solutions to emerge if there were customer demand. CS and CPS today represent a useful means of opening up the network to competition – they represent a way to get past the incumbent’s switch. without incurring the high up-front cost of the local loop (e.

124

Final Report: The Future of IP Interconnection

5.3.4 Related threats to competition
It is worth noting that literal barriers to interconnection at the IP and the service level do not represent the only potential threats to competition as networks evolve in the direction of NGN. The Commission’s 2003 study207, and also Ofcom’s consultations, have explored the possibility that emerging IP-based networks might introduce new “choke points”, new threats to competition. The 2003 study saw location information, as well as customer identity systems, as particularly relevant. As the 2003 study observed, “Essentially, NGN has the potential of providing a more open and competitive service environment. Paradoxically, this potential may also lead to additional sources of market power. Whereas traditionally, all elements of service creation were controlled by a single operator, the NGN environment enables many of these elements to be provided competitively. Where such elements have to be chained together in order to create an end-user service, it follows that control over any single element in the chain would provide control over the whole chain. … the battle for market power will be fought on many different fronts.”

5.4

Different market players – different evolutionary paths to NGN

Different service providers, in different countries or competing within the same country, will follow different migration strategies towards NGN. This relates both to the speed of deployment and to the specific architectural and topological changes that will take place in their respective networks. These distinct NGN adoption strategies on the part of different service providers will have an impact on competition and market structure. Service provider migration plans will tend to reflect (1) the nature of their current preNGN business model, and (2) the company’s immediate business realities. Service providers of many different types are all expected to eventually migrate to IP-based NGNs, but not necessarily in the same way or at the same pace. In each case, there is a path dependency: the route that is taken and the final destination both depend somewhat on the point of departure. NGN migration plans are likely to be somewhat different for: • • • • • Fixed incumbents, Competitive fixed operators, Mobile Operators, Internet Service Providers (ISPs), Cable television operators that also offer Internet access and/or telephony services.

207 See Cullen/Devoteam (2003). See also Elixmann et al. (2007).

Final Report: The Future of IP Interconnection

125

5.4.1 Historically a telco, or historically IP-based?
A fixed incumbent without a significant mobile service might tend to favour a rapid NGN migration. They conceivably might not see a compelling need for IMS; however, we expect that most fixed incumbents will nonetheless include ETSI IMS in their NGN migration plans. Interviewees felt that the migration phases might be different across different countries and carriers; however, they consistently saw the ultimate goal of the transition as an integrated NGN/IMS solution. BT was initially unique in proposing a true “fork lift upgrade”, where the traditional PSTN was to be phased out altogether in a small number of years; however, we are seeing similar overtures from KPN, and we can expect to see more from other incumbent operators over time. Fixed incumbents, with or without mobile operations, will tend to replace a centralized PSTN switching environment with a still-somewhat-centralized IP-based server environment. The initial objective will tend to be to change the underlying technology base over which they offer their existing palette of services. To date, the primary drivers seem to be lowering unit costs and improving fixed-mobile integration, with somewhat less emphasis on deploying new services. (The migration to video is a driver, but this migration does not uniquely depend on NGN.) ISPs and cable operators are already evolving in the direction of NGN, but at a somewhat relaxed pace. They already have an IP-based infrastructure, so the changes in their case are gradual and evolutionary. Their evolutionary path tends to be distributed rather than centralized.

5.4.2 Incumbent or competitive entrant?
Competitive fixed operators and non-incumbent mobile operators will tend to follow paths somewhat similar to those of their incumbent counterparts in the countries in which they operate; however, some appear to be under less pressure than the incumbents to migrate their networks. As with incumbents, they will attempt to inject an IP transport beneath their existing services. Mobile operators are likely to emphasize IMS-NGN, while fixed-only competitive operators might not see so strong a need for IMS.

5.4.3 Fixed, mobile or both?
Typical fixed incumbents that also have substantial mobile operations will tend to have a strong preference for an integrated IMS-NGN solution, and might tend to spend more time on the transition due (in part) to the complexity of integrating their fixed and mobile operations.

126

Final Report: The Future of IP Interconnection

5.4.4 Implications of the different migration paths
Among these various plans, some key distinctions are evident: • The degree to which the service provider already possesses SMP that it might seek to leverage into related markets (for example, last mile market power that it might seek to leverage into the market for applications over the NGN), The speed with which the transition takes place, The speed with which the traditional PSTN disappears, especially for the wired incumbent, The degree of centralization or decentralization of the resulting infrastructure, Whether the service-provider is already primarily IP-based, and Whether the transition to NGN is likely to include IMS.

• •

• • •

Firms that are primarily telephony providers are more likely to evolve to centralized NGNs, while IP-based firms and cable operators are more likely to adopt a more gradual evolution and a more decentralized service architecture. The presence or absence of SMP is a central theme in European regulation, and this is likely to continue to be the case as networks evolve to NGNs. The evolution may change the character of market power, but is unlikely to eliminate concerns with SMP. Last mile bottlenecks are likely to remain for some time, as will the call termination monopoly; moreover, new bottlenecks might emerge in the upper layers of the network, even as other SMP manifestations may be ameliorated by new forms of competition. SMP shapes the migration path, and also conditions the corresponding regulatory response. SMP thus represents a common thread through our assessment of regulatory needs during the transition to NGN, which follows in Section 5.5.

5.5

Regulation and the transition period to VoIP and NGN

The period of transition to NGN raises many vexing problems for regulators. In many respects, the transition period is likely to be even more difficult than the steady state period that is likely to follow transition.

5.5.1 Traditional remedies versus new remedies
A first question that must be addressed is how long existing regulatory remedies should be maintained. Regulators presumably have to strike a balance. If existing pre-NGN remedies are prematurely lifted, there is the risk that competition as a whole will run

These issues play out a bit differently for interconnection. traffic is often delivered using ATM as a level 2 transport mechanism. three or four POIs should be technically and . 5. For most European Member States.Final Report: The Future of IP Interconnection 127 aground. very few POIs are required. the operator that provides the IP-based transport service to the end user will not want the operator that provides the physical last-mile IP access to be visible in the IP-level routing. should be) implemented using IP-based interconnection. At the same time. still enabling the service provider to offer an apparently end-to-end service to the end user. the regulator potentially inhibits network modernisation and promotes inefficiency. In the case of access. if the regulator appears intent on keeping previous remedies in effect indefinitely. In the case of interconnection. thus avoiding this problem. they will not necessarily be implemented using the same points of interconnection. Interconnection with the incumbent will presumably be implemented by means of IP-based peering. the United States). it is quite possible that the traffic will be “tunnelled”. In the NGN world.5. For peering-based interconnection. tomorrow. however. To the extent that the incumbent operator unilaterally reduces the number of Points of Interconnection (PoIs) in order to avoid needless cost. Most operators will want a minimum of two or three POIs for reliability and redundancy. both access and interconnection will be (or rather. both providers will necessarily be visible in the end-toend routing.g. In a large country (e. or alternatively perhaps by a transit relationship with the incumbent or with some other service provider that has a peering relationship with the incumbent.2 Number of Points of Interconnection (PoI) A particularly difficult NGN migration challenge flows from the positive opportunity to simplify and collapse the network into a smaller number of geographic locations. Today. without making the last-mile access provider visible in the end-to-end IP routing. but even there the largest IP-based operators historically have had on the order of ten POIs for purposes of interconnection. propagation delay due to the speed of light is such that a somewhat larger number might be desired. who should bear the cost of stranded investment associated with competitor build-outs to interconnect locations that are no longer serviced? What kind of notice and consultation mechanisms are appropriate? Much of the discussion to date of the reduction in the number of PoIs as a result of migration to NGN has been in terms of network access.

208 By contrast. not only is there a need to infuse substantial capital. this balance involves proper consultation and notice with competitors. Most regulators appear to be striving to strike some balance so as to preserve competition overall. possibly coupled with obligations provide reasonable notice prior to eliminating a PoI. This substantial reduction raises the concern that most or all competitive operators are likely to have substantial stranded investments in interconnection circuits and equipment to PoIs to the incumbent that will no longer exist in the incumbent’s new network.128 Final Report: The Future of IP Interconnection economically sufficient for purposes of interconnection. a key factor in their permissible rate of return on capital. operating expense should be lower than with current networks. Balancing risk considerations against operational expense that increases before it declines seems to represent a delicate balancing act. The reduction in the number of POIs for interconnection could potentially be even greater (in percentage terms) than the reduction in the number of POIs for access. including Ofcom in the UK. then no conventional points of interconnection at all to the incumbent are necessary. It is clear that the incumbent operators need to have the ability and the flexibility to modernize their respective networks. and thousands for access. in the near term. a large Member State today might have hundreds of PoIs for interconnection. 209 See Ofcom (2005) and OPTA (2006). A number of NRAs have looked at this problem. however. Often.209 5.3 The changing average incremental cost of service during the transition period The migration to NGN poses many challenges. The roller coaster ride of operating expense potentially represents serious challenges.210 which is substantial during the migration. 210 Risk was reflected in Ofcom’s calculation of BT’s Weighted Average Cost of Capital. .5. The NRA needs to consider risk. It is equally clear that the regulator cannot be in the business of protecting individual competitors. 208 If an operator chooses to interconnect using a third-party transit arrangement. This poses serious challenges for the regulator. but there is also the likelihood that operational expense will actually be higher on a unitized basis due to the need to operate the old and new networks in parallel. In the long term.

211 Alternatively. are low.g. It is important to note that arbitrage is not necessarily a problem in and of itself. especially as regards simultaneous co-existence of PSTN-PSTN. the regulator’s first response should not be to suppress it. especially for the voice service. so as to put upward pressure on the prices of their competitors. irrationalities in the underlying pricing structure. or the expectation of arbitrage. instead. VoIP services from Skype) further complicates the picture. independent of the broadband network that carries the service. they might find it advantageous to keep termination prices high. It is conceivable that arbitrage. and to a lesser extent on mobile voice services.4 A risk of arbitrage? Interconnection arrangements during the period of transition from today’s network to the IP-based network of tomorrow pose special challenges. To the extent that interconnection arrangements reflect different prices. Fixed operators may possibly find it difficult to reconcile these low retail prices with current call termination fees. To the extent that the underlying costs of a third-party VoIP service.211 whether this will be the case for mobile is unclear.Final Report: The Future of IP Interconnection 129 5. If substantial arbitrage were to emerge during the transition to NGN. and VoIP-VoIP interconnection arrangements.5. . the low costs will lead to low retail prices for VoIP services that will place downward pressure on the price of conventional fixed voice services. opportunities for arbitrage are likely to be substantial. PSTN-VoIP. and partially correct for. the regulator should first try to understand (and if appropriate to remedy) the root causes of the dislocation that makes the arbitrage profitable. The emergence of independent third party providers of services (e. Opportunities for arbitrage usually reflect. will drive retail prices to much lower levels than we see today. This may imply the need to keep interconnection arrangements consistent for VoIP and for PSTN during the period of migration from the switched PSTN/PLMN to NGN.

6. Furthermore. Newer operators and cable operators may move quickly to IP-based interconnection. is likely to persist for the foreseeable future.130 Final Report: The Future of IP Interconnection 6 Conclusions our Conclusions. Concluded in Conclusions cont’d on page 133. This evolution will significantly change the value chain of electronic communication services. The evolution to IP-based NGNs will place increasing strain on current interconnection arrangements for traditional telecommunications networks. Voice will be just one of many applications carried on tomorrow’s IP-based networks. To the extent that call termination were driven to much lower rates prior to the migration to an IP-based NGN. It is probably not appropriate for regulators to force these rates to zero. This will be accompanied by significant changes in the value chain of electronic communication services. Key Concept Box 21: Conclusions The clear trend for conventional telecommunications networks is to evolve into IP-based Next Generation Networks (NGNs). Voice will be just one of many applications carried on tomorrow’s IP-based networks – it will no longer be in such a unique position. The migration of voice services to IP-based networks poses especially profound challenges for interconnection going forward. and will continue for some time to represent a large fraction of the revenue of network operators. even as its fraction of network traffic declines over time. but established incumbents and mobile operators may prefer to maintain traditional voice interconnection methods. which derives from the fact that only a single operator can complete a call to a given E. Nonetheless.164 telephone number. . it would avoid disincentives to migrate to new IP-based interconnection arrangements. because the implementation of QoS between network operators may depend on non-zero payments. the market power that these network operators possessed in the PSTN/PLMN world will not necessarily diminish when they migrate their networks to IP. even as its fraction of network traffic declines over time. voice will continue for some time to represent a large fraction of the revenue of network operators. The basic negotiated arrangements for peering and transit traffic appear to be reasonable and sustainable.1 The evolution of the market There is a clear trend for conventional switched PSTN/PLMN networks to evolve into IP-based NGNs. A major finding of this study is that the termination monopoly problem. The next section provides our This section provides Recommendations.

for British Telecom.Final Report: The Future of IP Interconnection 131 6. or could soon become the case. The mechanics could 212 This is the case. In either case. and eircom. to the extent that termination needs to remain regulated. or that make the telephone number irrelevant altogether.2 The call termination monopoly A major finding of this study is that the call termination monopoly is likely to persist for the foreseeable future. it becomes necessary to ensure that the access at IP level to that termination remains free from anticompetitive restrictions. there has been great interest in Europe (and elsewhere) in functionally separating fixed electronic communications network operators so as to simplify regulation. Telecom Italia. Previous studies have tended to focus on bottlenecks in the access network. For entities that are subject to some degree of functional separation. however. the migration to IP in and of itself will not. . The principle of technological neutrality would appear to apply with full force. the non-discrimination provisions are still necessary. In recent years. There are also examples in the cable world. but the VoIP-based telephony service were to remain with the retail service portion of the company. The termination monopoly derives from the fact that only a single operator can complete a call to a given E. however. if the relevant IP interconnection capabilities remain with the retail operation. Technology that makes it possible for multiple services to complete voice services to a phone number. this was a somewhat unexpected finding.164 telephone number. Suppose that the IP interconnection facilities were effectively transferred to the access services portion of a functionally separated company. then the non-discrimination would be solely a matter for the retail operation. and would not be simplified by the functionalseparation. We mention this at the outset because it has large implications relative to our conclusions. call termination payments (meant to reflect the use of network facilities) would presumably have to flow to the retail telephony operation – some internal funds transfer would presumably be necessary from the retail voice service to the network operation in order to compensate the functionally separated access network. Working back from the termination monopoly. Notably. might change this in the very long term.212 it might be easier to enforce non-discrimination provisions (depending on the details of the separation). For the study team. It would then be easy to implement non-discrimination. and have paid only limited attention to the termination monopoly in an IP-based world. Experience has shown that the termination monopoly applies equally to fixed and mobile telephony. we find a need for regulatory obligations that might otherwise have been unnecessary. but the termination monopoly will persist as long as Europe remains on CPP/CPNP arrangements. It will change somewhat in character as voice telephony becomes de-coupled from the network. however.

213 See Katz/Shapiro (1985).3 Interconnection obligations for data This section deals with interconnection obligations. larger operators may have countervailing bargaining power to the extent that they can threaten to refuse to interconnect. We have seen no indications that their charges are problematic. they are likely to be highly motivated to interconnect with larger traditional operators.132 Final Report: The Future of IP Interconnection vary greatly from one network operator to the next depending on the nature of the functional separation.214 however. 215 See Dewenter/ Haucap (2005). those conditions have been met without regulatory intervention in most cases. A number of papers suggest that asymmetric regulation of termination (imposition of an obligation to interconnect without a cap on the termination fee) leads smaller players to assess even higher termination fees than they otherwise would. however. in the absence of a regulatory cap to termination fees. desirable.2.2. regulators need to be able to intervene where necessary if the minimal conditions are not met. To the extent that these service providers have small market share.213 Some Member States permit these third party operators to assess termination charges. the next deals with the closely related question of the level of charges for interconnection. and on the ultimate decision as to what should be basis for termination payments. As long as the IP-based network operator with greatest market power in a Member State has peering agreements under reasonable terms (which typically would reflect low or zero interconnection prices) with at least two or three major competitors. it is not necessary. In an IP-based world. . conversely.1. a combined system of peering and transit should generate a good market outcome. An open question is how best to address the terminating monopoly in the case of third party VoIP service providers who do not have a real network of their own. larger operators might not be motivated to interconnect with them in the absence of a regulatory obligation. however. as networks migrate to NGNs. nor even feasible for every IPbased network operator to have direct interconnection with every other. the theory suggests that they should be motivated to charge supracompetitive termination fees. 214 See Section 3.215 6. In the Internet. Small operators should be motivated to charge even higher termination fees than large ones.

Mobile operators are likely to try to maintain the current arrangements. desirable. As networks migrate to NGNs. and that no public policy intervention is required. At the same time. In most respects. a combined system of peering and transit arrangements should generate a good market outcome overall for IP data interconnection. It should not necessary be that all have direct physical access (to one another – IP transit arrangements should suffice. If each operator that has significant market power for termination were to have IP peering arrangements of good quality with several direct competitors (but not all) in a Member State. this should apply equally to the use of IP to access call termination facilities. including the likely exhaustion of IPv4 addresses in the 2010-2011 time frame. They depress mobile voice usage. They may also depress the mobile operator’s ARPU. nor even feasible for every IP-based network operator to have direct IP data interconnection with every other. because they are highly profitable. we think that the evolution of IP and NGN standards is proceeding as it should. and the possibility of improved guidance on quantitative Quality of Service (QoS) standards and measurement. and thus consumer welfare. as long as a system reflecting a mix . but also comparing the quality of IP interconnection that the operator with significant market power offers for call termination uses for its own IP-based services versus the quality of IP interconnection that it offers to competitors. Key Concept Box 22: Conclusions cont’d It is clear that current mobile call termination fees are much too high.Final Report: The Future of IP Interconnection 133 In principle. then one could reasonably expect that transit arrangements to other domestic competitors would be available at reasonable price and quality. A numbering of issues nonetheless warrant attention. but a number of Member States are implementing reasonable responses. As long as the IPbased network operator with greatest market power in a Member State has peering agreements at low or zero interconnection prices with at least two to three major competitors. The migration to NGN raises numerous transition questions. it is not necessary. In an IP-based world. and were not permitted to discriminate against those competitors in regard to the quality of IP provided to their voice services. # # # In an IP-based world. regulators need to be able to intervene where necessary to ensure that this continues to be the case. operators that have Significant Market Power with regard to call termination – which effectively includes all operators who terminate calls – can not be permitted to discriminate against competitors in regard to the quality of IP interconnection used to access their call termination facilities. This is more akin to equivalence of input than to non-discrimination. there could be a far greater number of market players. This suggests once again that it should not be necessary to mandate any-to-any IP peering interconnection. Discrimination in this sense means not only among the operator’s competitors. the growing complexity of IP routing tables due to an increasing level of multiple paths to connection to a given IP address.

we discussed the presence or absence of an obligation to interconnect. we do not see the need for a new remedy for IP interconnection in support of voice. it is possible that regulators would find such a system more difficult to monitor than a set of arrangements with any-to-any peering. we observe that the migration to IP introduces the possibility of introducing subtle. Based on the principle of proportionality. their obligations should be crafted so as to address the harmful effects of that market power but to otherwise impose as few constraints as possible on them. . terms or conditions substantially inferior to those which the SMP operator provides to itself. 6. we see the need to ensure that NRAs have any necessary authority to ensure that voice interconnection remains effective as networks migrate to an IP basis. First. these concerns are present whether voice traffic flows over a specific IP-based voice interconnection point. The new challenges derive primarily from the incorporation of voice into the IP environment. the NRA (or NCA) might take action ex post if an SMP operator were to fail to do so. This still does not necessarily seem to imply the need to impose non-discrimination obligations on all IP traffic. This might take the form of requiring ex ante that SMP operators not provide IP-based access to their voice services under quality.4 The level of charges among IP-based network operators In this section. it is appropriate to apply remedies where necessary to address market power. Rather.134 Final Report: The Future of IP Interconnection of IP peering and transit is effective. In terms of a policy response. not a regulatory decision. we consider the level of charges. we should make some technical observations. alternatively. thus raising the question of how deviations would be detected and demonstrated. 6. In principle. that should be a commercial decision. hard-to-detect impairments of service quality. in the previous section. For undertakings that have market power only in regard to call termination.4. however. Second. or over the same interconnection facilities as other IP data. Basic arrangements for peering and transit traffic appear to be reasonable and sustainable. Finally.1 The level of charges for IP data traffic The future level of interconnection charges for IP data are likely to have more in common with current Internet mechanisms than with current PSTN mechanisms.

The violations that are most worrisome are those that imply economic foreclosure: For example. especially at such time as existing switched voice interconnection is withdrawn. Traffic prioritization per se is not necessarily harmful to consumer welfare. regulators may need to have the authority to intervene for certain kinds of violation of neutrality. In terms of consumer welfare. Moreover. however.Final Report: The Future of IP Interconnection 135 IP-based operators who agree to peer usually (but not invariably) agree today not to charge one another for peering. 216 Including changes to Articles 20 and 22 of the Universal Service Directive. notably including the consumer broadband market such that most European end-users are able to choose among three or more providers. and secondarily on the QoS that has been requested (and delivered). some related to the security. Implementation of QoS between network operators probably depends on non-zero payments. there is a good likelihood that Network Neutrality will not turn into a noteworthy problem in Europe. a network operator blocks or degrades the quality of access to some website that competes with a website affiliated with the network. preventing the evolution of positive service differentiation. (This thought is in the spirit of a suggestion of Laffont and Tirole (2000). we do not see a need for comprehensive new regulation to address Network Neutrality challenges. an overly stringent obligation of non-discrimination obligations could do more harm than good. Most likely. they will charge for IP data traffic primarily based on traffic volumes. operators would be motivated to implement efficient Ramsey-Boiteux pricing structures [that is. As long as European regulators maintain competition in underlying markets. Nonetheless. as already noted in section 6. and the ability to mandate a minimum Quality of Service. There are many open and challenging practical questions regarding charging for QoS. and some to verification that the interconnection partner has in fact delivered the level of QoS that it committed.) Where fixed network operators bilaterally agree on charging for IP data traffic. or blocks access to an independent VoIP service provider who competes with the network operator’s own voice services. if they charge at all. European regulators already have a range of potential remedies that collectively should suffice. and the minor enhancements that the Commission proposed on 13 November 2007216 further strengthen the hand of the regulators. pricing structures that reflect the degree to which demand is elastic in response to price] themselves. it is desirable that wholesale interconnection payments be either low or zero. For now. which are best determined by the operators themselves. who proposed that if regulators were to set broad overall caps on profitability. it is probably not appropriate for regulators to force rates to zero.2. they will necessarily do so on the basis of something that both can measure and verify. . In any case. some to the risk that the system will be “gamed”. The one area that we have identified where possible new obligations might prove necessary is in the IP data interconnection used to support voice interconnection.

but these charges should not be materially different from those associated with similar QoS guarantees for other data traffic There are indications that this is already happening among third party VoIP service providers. one might hope that current wholesale voice-oriented call termination arrangements in the fixed network would break down. For all of these reasons. and in any case are more appropriately measured by the service provider than by the network operator. it is not clear that incumbent fixed operators would choose to evolve to such a model. and also among cable operators that offer VoIP services. independent third party VoIP service providers will put enormous pressure on retail prices.2 The level of charges for voice The likely evolution in the absence of regulatory intervention might be quite different for the fixed network than for the mobile network. More generally. and may in some cases also negatively impact the mobile operator’s ARPU and profitability.4. because their price will be driven by their real costs. In terms of consumer welfare. For the fixed network. The avoided usage represents a negative impact on consumer welfare. it is clear that current mobile call termination fees are much too high – as noted in Section 3. the voice service would clearly play a much smaller role than it does today in these charging arrangements for the fixed network. Minutes of voice telephony use in the fixed network are only weakly correlated with network cost. first. voice services drive only a small fraction of the cost of today’s networks.2. the mere presence of independent third party VoIP service providers calls into question the relevance of charging arrangements where the voice service seeks to recover the costs of the underlying multiple service network. traditional arrangements will be under progressively increasing pressure in the next few years. but also as a means of keeping costs higher for their competitors and thus reducing the downward pressure on their own retail prices. because the termination fees that they receive are so much larger than . not only as a direct source of revenue. that the mobile operators will be much more strongly motivated than their fixed counterparts to maintain the current arrangements. however.3. probably in favour or charging arrangements that track more closely to real average incremental costs. What should ideally happen for the mobile network is not very different than for the fixed network – one would hope that wholesale and retail prices would move much closer to actual cost. and this fraction will continue to decline. They may instead prefer to maintain current termination arrangements.2. There might be charges for the Quality of Service with which the underlying traffic were carried. Furthermore. they depress mobile voice usage.136 Final Report: The Future of IP Interconnection 6. It is even more difficult to predict what will happen for the mobile telephone network. and would reflect the declining role of voice in comparison with data over time. To the extent that voice termination fees are unrelated to real average incremental cost. In this scenario. What is different in the mobile environment is.

current interconnection arrangements will come under increasing strain as switched PSTN/PLMN networks evolve into IP-based NGNs.1. 6. • We return to these questions in Chapter 7 of this report. and How to address changes in the number of Points of Interconnection. they have far greater ability than their fixed counterparts to inhibit or at least to delay the deployment of third party VoIP services onto their mobile networks. however. as noted repeatedly in this report. what will happen in the absence of a public policy response is a more complex question. 217 See Section 5. necessitate serious consideration of a number of key challenges: • • How long current SMP obligations should remain in force. Lack of regulatory clarity risks delaying network evolution.4. When and how new SMP obligations tailored to the new IP-based environment should be put in place. We return to these points in Chapter 7. Again. the legal and philosophical basis for doing so requires careful consideration. Second. but established incumbents may prefer to maintain traditional voice interconnection methods until economic arrangements for IP-based interconnection have been clarified to their satisfaction. To the extent that call termination arrangements based on much lower fees were adopted prior to the migration to an IP-based NGN. Newer operators and cable operators may move quickly to IP-based interconnection. as previously noted.5 The migration to an IP-based NGN The migration to IP-based NGNs will. may have both the incentive and the ability to sustain current arrangements for some extended period of time. which deals with recommendations. it would avoid disincentives to migrate. Mobile operators. .Final Report: The Future of IP Interconnection 137 those that flow to the fixed operators.217 What should happen in the mobile environment is clear enough.

but we do not have a specific recommendation to offer to address the issue.6 Technical considerations in the evolution of the IP protocol In most respects. and that no public policy intervention is required. The gradual “bloat” of IP routing tables. as noted in Chapter 2. either in terms of research or in terms of active policy: • • • The likely exhaustion of IPv4 addresses in roughly the 2010-2011 time frame. and The possibility of improved guidance on quantitative Quality of Service (QoS) standards (delay. we have. between IP-based networks. and the techniques with which to measure them. and are therefore relevant to this study.138 Final Report: The Future of IP Interconnection 6. Our interviews also suggested concerns with the difficulties that researchers have in gaining access to data about traffic evolution in the Internet. These issues are reflected in our Recommendations in Chapter 7. . however. we think that the evolution of IP and NGN standards is proceeding as it should. but they certainly have the potential to disrupt IP interconnection. jitter and loss) to be used. identified several potential challenges or opportunities regarding the ongoing evolution of the Internet Protocol (IP) that potentially have direct relevance to the European Commission. The first two are not specifically interconnection issues.

the remedies already available under the European regulatory framework are working.6 reviews technical considerations.Final Report: The Future of IP Interconnection 139 7 Recommendations Based on our findings. sections 7. NRAs must be able to intervene if interconnection breaks down. this power to intervene must include the ability to cap the price of interconnection. however. while section 7. with or without payment – this is the ideal case.1 Interconnection obligations – IP traffic We do not advocate an interconnection obligation as regards IP data traffic in general.4 consider interconnection obligations and prices for voice. however. The one conspicuous challenge that might emerge going forward is the possibility that an ECN or ECS might have market power by virtue of network externalities rather than according to standard tests of market power in one of the markets susceptible to ex ante regulation. and the price of interconnection. Electronic Communication Network (ECN) operators that possess SMP. Entities without market power will generally choose to have good quality interconnection. Entities that have roughly comparable bargaining power will often voluntarily agree to interconnect with one another. Key recommendations are captured in the introductory text to each section. primarily related to network addressing and to QoS. and might be disinclined to offer fully effective interconnection in order .5 reviews the issue of Network Neutrality.1 and 7. To be effective. It is not necessary or appropriate for the NRA to mandate any-to-any IP interconnection. and we do not see a need to mandate any-to-any peering. NRAs should however be prepared to intervene where necessary to address market power. We distinguish between the obligation to interconnect. for example in regard to broadband access. Sections 7. might not be motivated to implement fully effective IP interconnection on their own. If intervention should prove to be required. and will continue to work. in which case application of the SMP remedies in the Access and Interconnection Directive (notably the Article 12 obligation to interconnect networks or network facilities) may be appropriate. 7.2 consider interconnection obligations and prices for data. in light of the possibility of a mix of peering and transit arrangements. especially where this is a manifestation of some form of market power. we make the following recommendations to the Commission for the near to intermediate term. Section 7.3 and 7.

rather. or the quality of the IP interconnection that will have become the basis for voice interconnection. 7. It is possible to degrade the quality an IP interconnection in subtle. the terms and conditions. then there needs to be some constraint on the price. As noted in Chapter 6. It is conceivable that this might lead to manipulation of the prices. we conclude that the call termination monopoly is surprisingly durable. but leave operators free to choose the structure of prices. This enables the operator to choose an appropriate RamseyBoiteux pricing structure (that is. This is not to suggest that a new remedy is needed. It might be sufficient to require that the overall quality of IP-based voice interconnection that an SMP voice operator provides to competitors cannot be unjustifiably inferior to that which it supplies to itself (a condition more akin to equivalence of input than to non-discrimination). it is possible that some operators would respond by attempting to capture economic rents from other operators in some other way. Article 5(1) of the Access and Interconnection Directive provides NRAs with a sufficient tool to take appropriate measures should this still somewhat hypothetical scenario emerge. otherwise the obligation is meaningless.4. It may be necessary for NRAs to take action from time to time (possibly ex post) to ensure that operators subject to an SMP remedy do not take actions that would render the IP-based voice service interconnection ineffective or unusable.3 Interconnection obligations – voice call termination If call termination fees were in fact either eliminated or greatly reduced. it means that NRAs must be empowered to ensure that standard interconnection remedies (updated to deal with new technology) are not subverted. 7. . difficult-to-detect ways.140 Final Report: The Future of IP Interconnection to exploit its market power. Cremer/Rey/Tirole (2000). it should address the aggregate level of prices. It is 218 See Katz/Shapiro (1985). however.2 Wholesale payments for IP traffic interconnection Should intervention related to the price of IP data interconnection prove to be required.2. even in the face of profound changes in the technology and the market.218 The European regulatory framework does not provide a comprehensive solution to interconnection problems in the absence of conventional SMP. as we recommend in section 7. and also makes it possible to apply differential network-to-network interconnection prices to reflect different levels of Quality of Service. Should interconnection obligations prove necessary as an SMP remedy. to implement a pricing structure that reflects the degree to which demand responds to price).

interconnection obligations are generally in place. which is a decision best left to the market. This could perhaps change in time. for example.S. ECSPs will generally be motivated to interconnect. but are also willing to charge somewhat lower termination fees when they do. . Inasmuch as both fixed and mobile operators possess SMP in regard to call termination. conditions. an ECSP with a large customer base could. Alternatively. 220 See Katz/Shapiro (1985). It is important to note that this should not be viewed as a new remedy. just as would be the case if some SMP operator attempted to vitiate the effectiveness of an SMP remedy in the existing PSTN world. widespread deployment of user ENUM might make it possible for the caller’s handset to automatically identify the most cost-effective way to reach an end-user. Even in countries such as the U. that use Bill and Keep arrangements (and thus avoid inflated call termination payments). In fact. The European regulatory framework already provides all necessary tools. In terms of willingness to interconnect. rather. it might be sufficient to deal with problems ex post. an operator that does not have a network (to speak of) could still possess termination monopoly power. NRAs have all the authority they need to impose interconnection obligations as appropriate. threaten to refuse to interconnect with an ECSP with a small customer base in order to try to extract higher termination payments. or to raise its rivals’ costs. ECSPs with small market shares have strong incentives to interconnect. however.219 but assuming that Europe continues to use CPP/CPNP arrangements it will not change in the near or intermediate term. and quality) relative to those that the SMP operator provides itself for its own voice services. To the extent that IP-based interconnection becomes the basis for all voice interconnection. network externalities come into play.164 telephone number. change this market power. ECSPs with large market shares have less incentive to interconnect. All of this is independent of the question of whether IP data traffic uses the same physical interface as voice.Final Report: The Future of IP Interconnection 141 likely to persist as long as only a single ECSP is able to complete calls to a single E. it may become necessary to mandate ex ante that a network operator with SMP for call termination (fixed or mobile) must make IP-based interconnection available to competitors for purposes of voice service interconnection on nondiscriminatory terms (in terms of price. In a totally unregulated world. 219 For example. but will prefer to charge even higher termination fees. in and of itself. it means that NRAs must have the ability to enforce existing remedies.220 Paradoxically. Cremer/Rey/Tirole (2000). The migration to NGN does not.

for the reasons previously noted. moreover.2.4 Interconnection payments – voice call termination Independent of the migration to NGN.2.4. we think that societal welfare would be substantially enhanced if mobile termination rates were much lower than they are today.142 Final Report: The Future of IP Interconnection 7.4 and 5. and preferably no higher than the rates that prevail today for fixed termination rates. termination fees are still much too high in our judgment. and it involves the fewest impediments to the evolution over time of interconnection arrangements as networks evolve to an IP basis. and (2) a glide path would be necessary in any case in order to enable operators to adjust their retail payment plans over time to match the changing wholesale arrangements. . nonetheless.3. this is not specifically an IP issue – it exists independent of IP. A reduction in termination fees. Efforts in the European Union since 2003 to float mobile termination fees down to cost-based levels have brought large improvements. or (3) by permitting negotiated termination fees subject to an obligation that the fees be reciprocal (the same in both directions) between each pair of interconnected (fixed or mobile) networks.1. • 221 See sections 3. or (2) by requiring all fixed and mobile operators to eliminate call termination fees altogether. it minimises economic distortions. but at the cost of a massive regulatory intervention. termination fees for mobile operators were unregulated. and were enormously too high. inasmuch as the current arrangements greatly complicate the migration to IP-based NGNs. Prior to 2003. because (1) it is not clear how it would be justified under the current regulatory framework. ideally zero. • In an ideal world. we would offer some observations. We nonetheless stop short of recommending the second option. so as to reasonably quickly achieve levels much lower than those that pertain today. but the migration to IP-based NGNs puts the issue in sharper focus. could be implemented by (1) accelerating the speed with which the maximum call termination rate declines from year to year under existing CPNP arrangements.221 Again. the second option would appear to offer many advantages in comparison to the other two. especially in mobile termination fees. Current arrangements result in mobile termination fees that are inefficiently high. Call termination arrangements in the current voice network are perhaps beyond the remit of our current study. The outright elimination of call termination fees is simple. We expand on the rationale at the end of this section.

We are not indicating exactly what the target levels for fixed and mobile voice termination should be. flat rate or “buckets of minutes” plans because we do not think that regulators need to decide. We note that zero is in many respects an ideal number. This corresponds to option 1. Wholesale arrangements do not dictate retail arrangements. whose costs are in effect assumed to represent a reasonable proxy for those of the rest of the industry. Those costs then propagate to many other 222 Under the authority provided by Article 19 of the Framework Directive. since this is far beyond the scope of this study. were mobile termination fees to be in the same range as current fixed termination fees.Final Report: The Future of IP Interconnection 143 • What we concretely recommend instead is that the Commisson mandate222 that fixed and mobile call termination rates “fast glide” to pre-specfied target levels over a predefined number of years (somewhere between three and five). Reduction in mobile termination rates would tend to lead to a substantial decrease in usage-based retail prices for mobile service. RPP. we think that it is not ideal. and allows for mid-course corrections if necessary.S. at the expense of a significantly smaller increase in the fixed portion of the retail price of mobile service (the “waterbed effect”). regulators should strongly prefer to apply any remedies that might be required solely at the wholesale level. rather. Even though U. and to substantially increased use (expressed in minutes of use per month) of mobile telephones. If wholesale call termination rates were to move to sufficiently low levels. the regulator (state public utility commissions with some support by the FCC) does not attempt to track average incremental cost for all operators. they monitor costs only for fixed incumbents. There is a strong argument that this lowering of mobile termination rates should be implemented before the migration to IP-based NGNs. except to the extent that high termination fees tend to force CPP plans with still higher retail price per minute.S. in the U. attempts by operators to preserve high rates would tend to distort the migration. Recall that. . to increased availability and consumer take-up of flat rate and “buckets of minutes” retail plans. experience suggests that the third option would eventually lead to a positive outcome. We also observe that.. Retail regulation would be imposed only as a last resort. the fixed and mobile service providers would be able to freely choose the most appropriate retail plans to offer based on their respective business strategies and on the needs of their respective customers. We have not addressed the question of whether regulators should prefer CPP. otherwise. for reasons noted above. • • • Consistent with the general philosophy of the European regulatory framework. This fast glide path gives the operators time to adjust their business plans. one could reasonably expect that flat rate retail plans would emerge that included off-net calls to mobile phones.

aAB = aBA.S. .S. and Atkinson/Barnekov (2000). 226 Notably Armstong and Wright (2007).4.225 We expect that some readers will be puzzled by our suggestion that termination rates for calls to fixed phones and to mobile should be set to levels that could be lower than cost. mobile-to-mobile interconnection agreements often are based on termination fees of zero as a purely voluntary outcome. the arrangements are difficult to understand (even for experienced U. It should not be viewed as sacrosanct.226 Given the ambiguities involved.-inspired system of reciprocity is less desirable for Europe than the fast glide path that we advocate here. practitioners). and ideally to zero. The analysis of two-sided platforms can also shed useful light on the problem.144 Final Report: The Future of IP Interconnection interconnection arrangements as a result of the obligations of reciprocity. 227 Littlechild (2006). We think that a U. on the other hand. i. 224 See Marcus (2004b). We would start by noting that there is no perfect rate. A number of forward-looking theorists in the U. Valletti and Houpis (2005). and Littlechild (2006).S. Regulation of termination charges is a Sisyphean task. arguably far better than their European counterparts. have been arguing for years for a migration from their current system to a comprehensive system of pure Bill and Keep arrangements. and also Valletti/Cambini (2005). B and C. puts it this way: “The idea of an ‘optimal price’ is a chimera: there are as many optimal prices as there are different specifications of the underlying economic model.224 A migration to reciprocity in the European Union could easily multiply these start-up “growth pains” by 27 Member States.”227 223 Reciprocity is different from symmetry.223 Interestingly. who is not only a prominent economist but also an experienced regulator. requires reciprocity.e. Aside from that. Littlechild. but none require reciprocity between fixed and mobile operators. Current U. The notion that the optimal rate should reflect marginal cost (which is not really knowable in any case) was based on earlier theory that ignored network externalities and many other factors. Some European Member States require symmetry. arrangements are complex. For networks A.S. 225 See DeGraba (2000).1. and would probably reflect network externalities. With the consultancy fees available there is no likelihood of economists running out of ideas or agreeing with each other. With the cross-subsidies and economic rents at stake the interested parties can be expected to continue to devise new models and to argue their corner. would imply that aAC = aBC. the regulators challenge is enormous. and possibly other factors as well. A number of quite credible newer studies suggest that the ideal rate would be quite different. but not symmetry. The U. and should therefore play no role whatsoever in the network operator’s recovery of the costs of the call (not even through the monthly subscription fee). reciprocity means that the termination fee from network A to network B is the same as the termination fee from network B to network A. but it took years of fine tuning to get them to that point. benefits to the call recipient. Current arrangements reflect the unrealistic assumption that the call recipient derives no benefit from the call.S. See Section 3. Symmetry. They work reasonably well today. and have odd side effects.

They impose the total cost of the network on a voice service that will going forward be only one service out of many delivered over that network.1. especially where there is a risk that an ECN operator or ECSP might leverage market power into an otherwise competitive upstream or downstream segment. Regulators should however be vigilant as regards possible anticompetitive discrimination. it is clear that current rates are too high. the existing regulatory framework for electronic communications probably provides adequate tools. In general. 229 At the same time. 7. Current asymmetric termination arrangements were presumably intended to reflect very different costs of fixed versus mobile. the current termination arrangements represent an economic distortion for all of the reasons noted in previous chapters. If the minimum quality of service were set too high. For reasons that we have made abundantly in earlier chapters. regulators will need to take care. The true cost disparity is vastly less than the current 9:1 ratio of mobile termination to fixed (double transit) termination that is present on average in current European termination rates. aside from that. there can be a wrong rate.Final Report: The Future of IP Interconnection 145 Even if there is no right termination rate.5 Network neutrality In regard to network neutrality.228 It seems to us that mobile termination fees that are no greater than fixed would be significantly closer to real costs than are the current highly asymmetric arrangements. however.229 NRAs and NCAs need to be prepared to address wilful deviations from network neutrality. especially where an element of economic foreclosure appears to be present.4. the model totally disregards the substantial value to the party that receives the call. we do not advocate major regulatory initiatives at this time. and competition law provides additional mechanisms. 228 See EU-Commission (2007). we do not see the need for new regulatory remedies. and that they are impacting welfare by depressing use and by introducing economic distortions into the relationship between the mobile network and the fixed. however. page 28. they tend to place the cost burden on a service that represents only a small fraction of capacity demands on the network going forward. and would avoid many of the distortions present in the current system. We see merit in the use of Articles 20 and 22 of the Universal Service Directive to require ECSPs to document their practices as regards blocking access to services or degrading the quality of access to services. it could impede competitive entry. this is a much greater disparity than is present in other regions of the world. There may also be merit in enabling NRAs to mandate a minimum quality of service. Aside from that. we believe that actual costs of mobile service are considerably less than those implied by current termination rates (see Section 5. Based on ITU research. . Annex II. Moreover.2).

however. which (after several false alarms) now seems to be quite likely in the 2010-2011 time frame. harmful leveraging is potentially actionable by the National Competition Authority (NCA) ex post as a competition law violation. and with defined tools and mechanisms for measuring these attributes. we could merit in the Commission’s other proposals in this area. NRAs could mandate (pursuant to their authority under Articles 20 and 22 of the Universal Service Directive) that ECN operators and/or ECSPs provide consumers with a public statement as to the circumstances under which they would intentionally block access or degrade the quality of access to a site or a service (for example. we think that the kind of QoS standards reflected in MIT’s White Paper represent a promising direction as regards the ongoing evolution of QoS between providers. First. The one area where preventive regulatory measures should be considered is in ensuring that consumers have the information they need to make informed choices. (3) research into the longer term issue of unbounded growth (“bloat”) of the BGP global routing table. and (4) research into technologies that could over time ameliorate the termination monopoly. Further. could help to simplify negotiations among network operators and thus to reduce the economic 230 The NRA should bear in mind. . jitter and loss. and for measurement of compliance to those classes. technology and standards appear to be evolving satisfactorily. it is not altogether clear what specific powers Article 5 confers on the NRA. such as user ENUM (rather than carrier ENUM) or techniques to enable multiple operators to serve calls placed to a single telephone number. that Article 5 needs to be used with caution and restraint. we see benefit in increased policy focus on a number of emerging issues regarding the technical evolution of the IP protocol. and no public policy response is required. Four areas where policy initiatives may be appropriate are (1) considering techniques to address or ameliorate the consequences of the exhaustion of IPv4 addresses. including the ability for NRAs to impose carefully crafted minimum quality standards on SMP operators.6.6 Technical considerations in the evolution of the IP protocol In most respects. Aside from that. inasmuch as it is not linked to a finding of SMP.146 Final Report: The Future of IP Interconnection NRAs already have substantial authority to address problematic deviations from Network Neutrality. and the NRA has the additional ability to apply Article 5(1) of the Access and Interconnection Directive. Non-discrimination obligations are already in place on most incumbents.230 Furthermore. (2) continued research and standards development in regard to classes of service for QoS. to block denial-of-service attacks). A standardized set of QoS classes with defined quantitative targets for delay. and thus well in advance of the full blown deployment of IPv6. 7. Based on the concerns raised in Section 6.

. The European Union. Research could be helpful. we think that increased research focus is required on the longer term issue of the steady growth in the size and complexity of the IP routing table. Increased research focus in this area merits consideration. not to ameliorate it. 231 The analogy to secondary markets for spectrum is strong. but also into the policy dimensions of what is likely to happen as IPv4 addresses become scarce. to its credit. has been advocating IPv6 migration for many years. but it nonetheless now appears that IPv4 addresses are likely to exhaust well before the IPv6 transition is complete. we think that we have finally reached a point where IPv4 address exhaustion represents an immediate challenge. not only into technical aspects of IPv6 deployment. The migration to IPv6 is likely to exacerbate the problem.231 thus potentially buying several years in order to enable the IPv6 transition to complete. Finally. Second. secondary markets would likely move large quantities of IPv4 addresses to higher valued use.Final Report: The Future of IP Interconnection 147 transaction costs that have slowed take-up to date of differentiated QoS between IPbased networks to date. One particular aspect that has not been looked at sufficiently to date is the possible use of secondary markets for IPv4 addresses – by forcing firms that currently hold large blocks of inefficiently used IPv4 addresses to confront the opportunity costs associated with holding their address blocks rather than selling them.

Jean (2000) : Connectivity in the Commercial Internet. pp. Vol.148 Final Report: The Future of IP Interconnection References Armstrong. 48. Vol. MPRA Paper number 4858. in: Economic Journal. pp. available at http://mpra. 185197.pdf. 14. in: International Journal of Communication. 433-472. John (2007): Net Neutrality: The Technical Side of the Debate . Crémer. Achim/ Eickers. Antonio (2007): Mobile Termination Cost Model for Australia. Brinkmann Michael/ Hackbarth.C. Charles/ Henseler-Unger.European Journal of Law and Economics. Version 6 (IPv6) Specification. 1–40. 2006. Mark/ Wright. Brussels. (2000): Bill and Keep at the Central Office as the Efficient Interconnection Regime. Report for the European Commission. 2003. 1. Harald (2006): Abschlussbericht der Projektgruppe Rahmenbedingungen der Zusammenschaltung IP-basierter Netze. D. in:. No. Opinion Paper. Iris/ Huber. Report for the Australian Competition and Consumer Commission. Vol.uni-meunchen. Carter. Eckhard/ Stöber. Armstrong. Karl-Heinz (2003): How Mobile Termination Charges Shape the Dynamics of the Telecom Sector. Werner/ Neumann.com/application/pdf/en/us/guest/netsol/ns537/c654/cdccont_0900aecd8 06a81aa. 2. Michael/ Wright. Vol. 22. Cullen International/ Devoteam (2003): Regulatory Implications of the Introduction of Next Generation Networks and other New Developments in Electronic Communications. Jacques/ Rey. DeGraba. 2006-2011. Ronald H. Vol.A White Paper. Detecon (2007): NGN and IMS. Warwick. in: OPP Working Paper Series. Cisco (2007): Global IP Traffic Forecast and Methodology. Bonn.ub. FCC. Final Report. 2003. 108. Julian (1999): Interconnection in Network Industries. San Jose. Patrick. Bad Honnef. Deering. pp. R. in: Journal of Law and Economics. Martin/ Le Blanc. RFC 2460. Carter. Julian (2007): Mobile Call Termination. CA. CA. 1998. Mark (1998): Network Interconnection in Telecommunications.cisco./ Hinden. Vol. . Bonn. 1-25. Bernd/ Knauer. in: Journal of Industrial Economics. 2007. in: Review of Industrial Organization. Paris. Coase. Bad Honnef. Matthias/ Lennertz. Peer/ Kurth. Gerd/ Eilers. pp. Washington. 27-46. pp. Julius (2005): The Effects of Regulating Mobile Termination Rates for Asymmetric Networks. in: Review of Industrial Organisation. 33. 20. 2007. Vol. 2000. Crowcroft. 567-579. Gilles/ Neumann. available at http://www. Klaus D. Oliver/ Cave. S. (1959): The Federal Communication Commissions. Dewenter. (1998): Internet Protocol. Berg. Karl-Heinz/ Portilla Figueras. Michael/ Wright. pp. Patrick/ Tirole. Karl-Heinz/ Spoerr. Sunnyvale. Horst/ Neumann. pp. Julian (2003): Asymmetric Network Interconnection. Tomas/ Fränkl. Bomsel. Dragan/ Neu.de/4858/./ Ilic. Ralf/ Haucap. 545–564.

Brussels. EU-Commission (2006): Commission Staff Working Document on Relevant Product and Service Markets within the Electronic Communications Sector Susceptible to ex ante Regulation in Accordance with Directive 2002/21/EC of the European Parliament and of the Council on a Common Regulatory Framework for Electronic Communication Networks and Services (Second edition). Washington. Lixin (2000): On inferring autonomous system relationships in the Internet. L 344/65.C. Berger and W.C. EU-Commission (2005): Review of the Scope of Universal Service in Accordance with Article 15 of Directive 2002/22/EC. European Regulators Group. Arlington./ Marcus. Annex Vol. Antonio/ Hackbarth. ERG 07 (09). Genakos. Brussels. EU-Commission (2007): 12th Report on the Implementation of the Telecommunications Regulatory Package–2006. Marc (2007): Az újgenerációs hálózatok szabályozási kérdései (NGN). 8 May 2003. P. 2007/879/EC. Bad Honnef. in: the RAND Journal of Economics. 2000. ERG (2007a): Final Report on IP Interconnection. 2007. FCC (2006): Annual Report and Analysis of Competitive Market Conditions With Respect to Commercial Mobile Services. SEC(2006) 837. FCC (2007): Industry Analysis and Technology Division Wireline Competition Bureau-HighSpeed Services for Internet Access: Status as of June 30. paper presented at the TPRC conference. 2007. Gao.. Garth (1985): Standardization. Incentives and Implications for Policy. and Innovation. Annex Vol. Study for the Hungarian Regulator. 2003/311/EC. ERG (2007b): Mobile Termination Rates Updated Snapshot. Brussels: Commission Staff Working Document. D. Vol. Peter/ Papai. Bauer. Joseph / Saloner. EU-Commission (2003): Commission Recommendation of 11 February 2003 on Relevant Product and Service Markets within the Electronic Communications Sector Susceptible to ex ante Regulation in Accordance with Directive 2002/21/EC of the European Parliament and of the Council on a Common Regulatory Framework for Electronic Communication Networks and Services. Compatibility. Lehr (2007): Complexity of Internet Interconnections: Technology. ERG 07 (27). European Regulators Group. 2007. Dieter/ Portilla Figueras. Christos/Valletti. EU-Commission (2007b): Commission Recommendation of 17 December 2007 on relevant product and service markets within the electronic communications sector susceptible to ex ante regulation in accordance with Directive 2002/21/EC of the European Parliament and of the Council on a common regulatory framework for electronic communications networks and services.. A. 16. Brussels. Budapest. COM (2004) 759 Final. in: Proceedings of the IEEE Global Internet Symposium. Faratin. Brussels. L 114/45. Gilmore. Washington. EU-Commission (2004): 10th Report on the Implementation of the Telecommunications Regulatory Package – 2004. Zoltan/ Scanlan. D. 18 December 2007. 2. 70-83. pp. S. 2007. Scott/ Nagy. 2007. Clark.. Klaus-D. J. 2007. 2. Brussels. Brussels. COM (2007) 155. 28 June 2006.Final Report: The Future of IP Interconnection 149 Elixmann. 2006. D. Tommaso (2007): Testing the “Waterbed” Effect in Mobile Telephony. Farrell. VA. P. Brussels. . 2006.

2004. Hermalin. Benjamin E. Papers and Proceedings of the Hundred . Jean (2000) On the "Receiver-Pays" Principle. March 1929. Vol. Stamatios V. Vol. Horrocks. Hotelling. Kartalopoulos . St. electronically available under: http://www. Jean (2000) : Competition in Telecommunications. Vol. Harold (1929): Stability in Competition. in: the RAND Journal of Economics. 423-448./ Vu. and Compatibility. in: the RAND Journal of Economics./ Shaprio. Bad Honnef. pp. J. Patrick/ Tirole. Klaus-D. pp. in: Computer Telephony. 125–145. Haucap. David (1997): The Rise of the Stupid Network. Vol. in: The American Economic Review. 4157. 1-37 Laffont. Jean (2003): Internet Interconnection and the Off-Net-Cost Pricing Principle./ Katz. Vol. pp. Katz. in: IEEE Communications Magazine. 91. pp.0 Communities. Robert/ Meckel. 2005. Paper presented at the ITS Conference. Kinder. Jean-Jacques/ Tirole. Dick (2006): IMS Based NGN Architecture and its Application. 85-110. Katarina (2006): Overview of Business Models for Web 2. Isenberg. Laffont. Patrick/ Tirole. Jean (1998b): Network Competition II : Price Discrimination. 8. 16-26. Laffont. Santander. Kruse. Scott/ Rey. 2000./ Kulenkampff. Paper Presented at the 12th CEPT Conference in Barcelona. 75. J. Roman/ Martignoni. Patrick/ Tirole. Jean-Jacques/ Rey. Jeon.pdf Knight. pp. 43.150 Final Report: The Future of IP Interconnection Hackbarth. Julius/ Marcus. (2004): Sender or Receiver: Who Should Pay to Exchange an Electronic Message?. Patrick/ Tirole. 2006.IMS Overview and the Unified Carrier Network. Jean (1998a): Network Competition I: Overview and Nondiscriminatory Pricing. MIT Press. pp. Study for the Bundesnetzagentur. Carl (1985): Network Externalities. pp. 29. Vol. pp.iec. 15-16. Laffont. 29. in: Rand Journal of Economics. Vol. J. Gabriele (2006): Technical Aspects of Interconnection in IP based Networks with Particular Focus on VoIP. Dale W. (2004): “Next Generation SONET/SDH Voice and Data”. 38-56. Gallen. Vol. Doh-Shin/ Laffont. John (2005): The NGN vs. Scott/ Rey.org/newsletter/sept06_2/analyst_corner. Jean-Jacques/ Marcus. Khuong (2007): Information Technology and the World Growth Resurgence. 35. WileyIEEE Press. in : The American Economic Review. 424-440. in: The Economic Journal. Kobe. Neil (2005): IMS–IP Multimedia Subsystem . in: German Economic Review. Vol. pp. Paper Presented at the ITU-T Workshop NGN and its Transport Networks. Jean (2001) : Internet Peering. Istanbul. April 2006. Jean-Jacques/ Marcus. Jorgenson. Jean-Jacques/ Tirole. 35. Miriam/ Stanoevska-Slabeva. Jörn (2007): Mobile Termination Carrier Selection instead of Regulation. 370-390. 2006. Michael L. Michael L. 2007. Laffont. Competition. in: Rand Journal of Economics. in: RAND Journal of Economics. Scott (2005): Why Regulate? Lessons from New Zealand. 34. the Internet and a New Model for Interconnection. Jean-Jacques/ Rey. pp. Högg.

Carpet. J. GSR Discussion Paper. J. 2007. pp.) The Internet Upheaval.pdf. Noam. Scott (2005a): Is the U.S.communications1b. Padmanabhan (2000): Competitive Effects of Internet Peering Policies. Bad Honnef. 287291 Littlechild. Daniel C. August 2006. 2006. Eli M (September 2005): P2P as a Market Catalyst. TPRC. January 2005. Marcus.Final Report: The Future of IP Interconnection 151 Thirteenth Annual Meeting of the American Economic Association (May 2001). Benjamin (eds. Bad Honnef. Scott (2006a): Framework for Interconnection of IP-Based Networks. 1991. Accounting Systems and Interconnection Regimes in the USA and the UK. Scott (2004b): Call Termination Fees: The U. pp. J.umn. Scott (2006b): Interconnection in an NGN Environment.umich. Addison Wesley. J. Paul/ Mitchell. Marcus. in: Journal on Telecommunications and High Technology Law.S. J. Dubai. Eli M. and the UK. Odlyzko. available at: http://www. J. Marcus. Andrew (2003): Internet Traffic Growth: Sources and Implications. Scott (1999): Designing Wide Area Networks and Internetworks: A Practical Guide. pp. 1-18. Scott (2005b): Universal Service in a Changing World. Dancing to a Different Drummer? in: Communications &Strategies. 2003. Scott (2007): Interconnection in an IP-based NGN Environment. Vol. WIK Diskussionsbeitrag No. Theory and Practice. Scott (2006c): Voice over IP (VoIP) and Access to Emergency Services: A Comparison between the U. Presented at the ITU Global Symposium for Regulators. in: IEEE Communications Magazine. Paper Presented at the 4th ZEW Conference on the Economics of Information and Communication Technologies. Cambridge University Press. VA.. Study for the Bundesnetzagentur.edu/~odlyzko/doc/history. Ingo (1991): Telecommunications Pricing. Bridger/ Srinagesh. Minneapolis. p. 175-195. MIT Press. Scott (2004a): Evolving Core Capabilities of the Internet. (2006) Mobile Termination Charges: Calling Party Pays versus Receiving Party Pays. Marcus. Marcus. Paper Presented at the TPRC. .si. available at: http://www. available at http://web. 30.pdf. in global perspective. AT&T Labs – Research. Vol. Marcus. Bridger/ Vogelsang. J. 2004. Marcus.comsoc. University of Minnesota. in: IEEE Communications Magazine. MIT Press. (2001): Interconnecting the Network of Networks.org/livepubs/ci1/public/2006/aug/cireg.H.comsoc. (2003): Explaining Internet Connectivity: Voluntary Interconnection Among Commercial Internet Service Providers. 3. pp. Mitchell. Mannheim.S. 2006. 60. Vol. 242-277. J. Andrew (2001): Internet Pricing and the History of Communications. 274. available at http://www. in: Telecommunications Policy. J. Mah. Marcus. Noam.org/ci1/Public/2005/jan/. Milgrom. 8-10. Odlyzko.edu/tprc/papers/2005/499/P2PMarketCatalyst.dtc. pp. in: Vogelsang. Ingo/ Compaine.html. 2003. 1999. Marcus. 123-163. Stephen C. MN.

Sam (2006): Internet Traffic Exchange: Market Developments and Measurements of Growth. Boston. Rohlfs. Ralf (2005) Ergebnisse der Nachfrageanalyse. Jean-Charles/ Tirole.itu. Toulouse. Bad Honnef. Moya/ Jung. Broadband Discrimination. Office of Communications. The Hague. 1. Ingo (2006): Abrechnungssysteme und Zusammenschaltungsregime ökonomischer Sicht. aus Wellenius. at: http://www. 2003. 430.html. Paul/ Mitchell. London. Bridger/ Paterson. Wu. in: Journal of Telecommunications and High Technology Law. Wu. Bad Honnef. 2003. 01/2007. Berlin. Reynolds. 2006. Prentice Hall. Jeffrey H. Elise (2007): Economic Study on IP Interworking: White Paper Prepared for the GSM Association. 2005. Uebele R. WIK (2005): Mobile Terminating Access Service: Network Externality and Ramsey Pricing Issues.int/osg/spu/ni/multimobile/legacyregulation/index. Study for the Bundesnetzagentur. OECD.1A.analysys. Dresden. Vogelsang. Jeffrey H. Bernd/ Schefczyk. 2002. (2003): Computer Networks. Jean (2004) : Two Sided Markets : An Overview. in : Institut d’Economie Industrielle working paper. SerCHo Consortium Deliverable D210. Peter/ Nicholls. 2007. (2007): Countervailing buyer power and mobile termination. George (2005): Mobile termination: what is the “right” charge?. at http://www. Study for the Bundeswirtschaftsministerium. Paul/ Dodd. 2nd Quarter 2001. World Bank Discussion Paper No. London. Washington. Vol. Tommaso M. Andrew S. Vol. 2005. Valletti. 2006. 2005. 2005. in: Alcatel Telecommunications Review. 141-179 . 2004. Tim (2007): Wireless Carterfone. in: International Journal of Communication. Paltridge. pp. Onafhankelijke Post en Telecommunicatie Autoriteit. Rohlfs. MIT Press. Paris. DSTI/ICCP/TISP(2005)11/Final. Rob/ Ball. Karlsruhe.com/usdownload/countervailing_report. Björn (2002): Closing the Gap in Access to Rural Communications – Chile 19952002. SerCHo Deliverable D220. pp. Rochet. (2001): Strategy for migrating Voice Networks to the Next Generation Architecture. Matthias/ Beckert./ Verhoeyen M. 2. (2003): Bandwagon Effects in High Technology Industries. Stuttgart.152 Final Report: The Future of IP Interconnection Ofcom (2005): Next Generation Networks: Further Consultations. 85-90. 389-426. 2007./Houpis. Tanenbaum. Telegeography (2007): Global Internet Geography Executive Summary. 2007. Astrid/ Waters. Michael (2007): Technologische und ökonomische Langfristperspektiven der Telekommunikation. A Consultancy Report to the Australian Competition & Consumer Commission in Relation to Optus’s and Vodafone’s Undertakings in Relation to the Domestic Digital Mobile Terminating Access Service. Pohler. OPTA (2006): The Business Case for Sub-Loop Unbundling in the Netherlands – Analysys Study.pdf. pp. Schäfer. Tim (2005): Network Neutrality. SerCHo Consortium with contributions by WIK-Consult (2005): State of the Art Analysis.

This implies (with isotropic calling patterns) that the entrant incurs an access charge deficit with the incumbent. Rey and Tirole. under certain conditions. Elsevier Publishers. In Armstrong’s interpretation. 1998a) that under symmetric conditions and reciprocal access pricing. including the primary literature on switched PSTN/PLMN interconnection. It is then clear that the entrant can only survive at lower prices. As a result. Armstrong.232 A body of useful work has emerged in the intervening years. Martin / Majumdar. Mark (2002): The Theory of Access Pricing and Interconnection. 545–564. 108. Sumit/ Vogelsang. then firms have no incentive to deviate from the collusive retail price because the gain in profits from undercutting the rival is just compensated by the increase in access payments needed for the increased number of calls going to the other network (and the reduced number coming from there). interconnection). Besides deriving a large number of results from the literature in a very coherent fashion. it can charge a monopoly price for termination.Final Report: The Future of IP Interconnection 153 Annex 1: Survey of recent published work on IP interconnection This annex summarizes key findings of articles and reports relevant to IP interconnection. this book chapter provides the first easily accessible treatment of Armstrong’s competitive bottleneck result. the firms can use access charges collusively to achieve maximal profits.): Handbook of Telecommunications Economics. As the other two it extensively treats one-way access and is somewhat shorter on two-way access (i. Armstrong terms the problem of monopoly over call termination one of competitive bottlenecks because it can happen under full competition between networks for subscribers.e. This paper is famous for deriving the result (independently discovered by Laffont. termination charges in small networks may even exceed monopoly prices. Armstrong. 232 Notably the WorldCom/MCI merger. finds value in cost-based (one-way) access prices. There are also several noteworthy papers that deal with technical aspects of IP peering and transit. In contrast to Laffont. in: Cave. Because the receiving network (usually) has a monopoly for termination to the subscriber being called. Economic analysis of IP interconnection was radically transformed by a series of high profile merger cases in the late Nineties. 295-384. Ingo (eds. at the same retail prices. the entrant’s market share would vanish. in: Economic Journal. Mark (1998): Network Interconnection in Telecommunications. even though traffic is balanced and therefore no net payments are being made. interpreting the entrant’s demand for access as its net demand (or access charge deficit). and the attempted WorldCom/Sprint merger.. In this case the access-pricing problem is similar to that of one-way access. Rey and Tirole. The papers on IP-based arrangements are best understood in juxtaposition with the economic literature on PSTN/PLMN interconnection. This is the most technical and most elegant of the three extensive surveys of interconnection pricing. Armstrong. pp. when access charges are set according to the collusive equilibrium rule. Vol. If callers only have average information about termination charges and do not know which network to which network the receiver subscribes. pp. the entrant will prefer lower access charges than the incumbent. . using the assumption that consumers prefer the incumbent over the entrant so that. Armstrong further considers a large incumbent and a small entrant.

89-105.fcc.edu/eps/mic/papers/0301/0301003. The paper develops a consistent. The latter is organized as club of academics who share the cost of their network. Where operators set a single rate. (2000): A Competitively Neutral Approach to Network Interconnection. pp. competitively neutral regulatory regime for interconnection between telecommunications networks.154 Final Report: The Future of IP Interconnection Armstrong. It shows that certain constellations and conditions are more favourable to peering than others. available at: http://econwpa. Badasyan. The paper that this default rule is competitively neutral and encourages efficient subscription and interconnection decisions. then combines the models. whether peering might threaten its existence. Two alternative modelling perspectives are used: the stable network is concluded being the complete network. Pio/ Wichmann. if the commercial networks gain considerably in size from peering and if their combined size is nevertheless not too large. They then consider reasons why mobile operators seem to set M2M rates equal to F2M. Julian (2007): Mobile Call Termination. It expands the discussion to incorporate a comprehensive model of call and network externalities.gov/osp/workingp. they suggest that it will tend to be intermediate between these values. The chances for such a threat will be larger if the cost savings from peering in the ARN are small.wustl. Subhadip (2003a): Private Peering Among Internet Backbone Providers.html. In the second part we discuss multilateral peering between commercial ISPs and an academic research network (ARN). thus lower than the monopoly price but still well above the efficient price. 1. Between commercial ISPs. a peering agreement decreases the profitability for a commercial ISP of using strategic pricing to win all members of the ARN as customers. Baake. but that they may prefer low M2M rates so as to moderate competition for customer acquisition. The analysis differs from previous work primarily in that it distinguishes the costs incremental to interconnection from those incremental to increased traffic volume. where all the backbone providers choose to peer with each other. Jay M. They find that mobile operators are unambiguously motivated to set high termination rates for F2M. Atkinson. Thorsten (1999): On the Economics of Internet Peering. and recover all remaining costs from their own customers. Mark/ Wright. where the backbone providers are connected to each other only through the national access points. while in the opposite direction large cost sensitiveness tends to increase the competitor’s network size./ Barnekov. however. absent a regulatory obligation. if there are a large number of operators. The analysis assumes a UK-like context. In this paper a model is developed in which Internet backbone providers decide on private peering agreements. where the fixed operators are distinct firms from the mobile operators. For the ARN is especially important. cost savings raise the chances of peering being profitable. in: Netnomics Vol. It provides two separate models of Fixed-to-Mobile (F2M) and Mobile-to-Mobile (M2M) call termination. This strategic pricing constitutes a stronger threat to the existence of the ARN who should therefore agree to a peering arrangement if that increases its members’ utility. It proposes a default bill and keep solution under which carriers split equally those costs that are solely incremental to interconnection. In the second part of the paper peering between an academic research network (ARN) and commercial ISPs is discussed. MPRA Paper Number 4858. while the efficient network is estimated being the one. comparing the benefits of private peering in relation to being connected only through national access points. the termination rate will tend to be close to the monopoly price. However. The article discusses economic rationales behind peering decisions in the Internet.pdf. Narine/ Chakrabarti. This recent paper expands on previous analytical models of call termination. available at http://www. Christopher C. . and to consider fixed-to-mobile call substitution.

Badasyan Narine/ Chakrabarti. 86. in: Economics Letters. As a consequence. The paper uses the framework of a simple model where two symmetric networks compete in nonlinear and discriminatory prices in the presence of call externalities. pp. in most cases companies would agree on above cost termination rates. In this paper a model of information exchange is developed to analyze competition between network operators. Thus. it is highlighted that from a social viewpoint bill and keep is welfare improving in comparison to cost-based access pricing.wustl. 1-25. Carter. The paper presents a simple game theoretical analysis of two Internet access providers who choose between peering and transit agreements. Reduced differentiation and the promotion of stiffer price competition are negative consequences. VA. 2003. Carlo/ Valletti. As evidence points to a positive propagation factor. unpublished paper. costbased access pricing can never be optimal from the social viewpoint. Subhadip (2005): A Simple Game Theoretic Analysis of Peering and Transit Contracting among Internet Access Providers. (2004): Access charges and quality choice in competing networks. which measures how return traffic is induced by sending calls. Furthermore. Their results suggest that providers do not necessarily exploit market power when refusing to peer. pp. while granting them full pricing flexibility due to the fact that optimal regulation is difficult and costly to achieve. Julian (1999): Interconnection in Network Industries.pdf. Using simulations. even if a regulator imposes marginal cost pricing at retail but regulates access charges only by requiring reciprocity. the networks can sustain monopoly profits by agreeing on high access charges (that increase marginal costs for each other). In this article a model with two backbones and an infinite number of ISPs connected to the backbones is developed. instead the joint profits are maximized under the transit arrangement. Instead. The role of the “propagation factor” is highlighted. Subhadip (2003b): Intra-backbone and Inter-backbone Peering Among Internet Service Providers. (2005): Information Exchange and Competition in Communications Networks. given the current system. under some circumstances. Paper Presented at the TPRC. who argue that cost oriented prices are likely to emerge in private negotiations. Narine/ Chakrabarti. It is argued that the market forces determine the decisions of peering and transit. in: Review of Industrial Organisation. Tommaso M. cost-based access pricing revisited. Their results are in contrast with the findings by Carter/Wright (2003). Vol.Final Report: The Future of IP Interconnection 155 Badasyan. Berger. Michael/ Wright. in: Information Economics and Policy. increasing demand and the ability to increase producer surplus by charging higher prices are highlighted on the positive side. Indeed. Carter and Wright recommend . The incentive and ability for tacit collusion. Also. Vol. It derives one robust result: once investments are endogenized. Cambini. Cambini. factors supporting or hindering peering are highlighted. available at: http://econwpa. it is shown that. The paper studies the impact of reciprocal access charges on the incentives to invest in upgraded networks. 107-112. if call externalities are taken into account. no need for regulation on interconnection policies is deduced. connectivity breakdowns are not interpreted as a major source of concern under a mild form of regulation (reciprocity). pp.edu/eps/mic/papers/0407/0407006. 391-409. negotiation over access charges leads to collusion that would be forbidden in the retail market. derived by Armstrong (1998) and LRT (1998) is maintained if brand loyalty leads to different sized networks. Carpet. Vol. the industry can self-regulate and achieve first-best allocations via negotiated access charges that internalize externalities. They suggest in case of operators with symmetric and balanced calling patterns to allow operators to negotiate over reciprocal access charges. Ulrich (2005): Bill-and-keep vs. 16. Carlo/ Valletti. March 2005. there is never an alignment between private and social interests over the setting of reciprocal interconnection charges. Tommaso M. It is shown that with two-part tarifs and discriminatory prices. 14. and thus. Quality improvements.

27-46. Economides. Jan H. 2000. 48. The use of two-part tariffs at retail thus could lead to access charges that would sustain welfare optimal usage charges downstream.gov/osp/workingp.. D. This suggestion also addresses the conjecture that incoming calls are. DeGraba’s COBAK (central office bill-and-keep) proposal restricts the bill-and-keep portion of a call to the termination of calls from the last central office to the called party. Vol. DeGraba. However. 20. Vol. Sumit/ Vogelsang. DeGraba suggests Bill and Keep with a twist. in Majumdar. Torben B (2003): Regulatory Implications of the Introduction of Next Generation Networks and Other New Developments in Electronic Communications. on average.156 Final Report: The Future of IP Interconnection regulation of access charges rather than regulation of retail prices. available at http://www. the larger one prefers a lower quality interconnection than the smaller one. Aurelie/ Guettler. Julian (2003): Asymmetric Network Interconnection. When two backbones interconnect. using regression analysis with country-level European data as a panel data set. while the sending network would be responsible for transport and switching until that point. while at a lower price. Letting the larger firm choose the interconnection charge but impose reciprocity is formally shown to be optimal by Carter and Wright in a non-cooperative model similar to L-R-T (1998a). However.html. 373-412. (2000) Bill and Keep at the Central Office as the Efficient Interconnection Regime. since./ Leeson. valued less than outgoing calls. No. but with a goodwill advantage for one of the networks. The paper develops intuitions about call termination fees based on the established theory. Jacques/ Rey. the suggestion no longer achieves the optimal outcome. The paper uses an extension of the Katz-Shapiro (1985) network model to analyze the strategies that would be used by a firm that dominates the backbone market. in: Journal of Industrial Economics.fcc. 185-197. the larger network provider will choose access charges equal to marginal costs of call termination. Dame. Vol. while it is undesirable because of the monopoly pricing outcome Carter. Morten/ Jensen. Patrick/ Tirole. Dewenter. It analyses the profitability of a “targeted degradation” strategy where a larger backbone would lower the quality of interconnection to its smaller rivals in turn. access charges only affect interconnection profits and not end-user profits. 2. Jean (2000) : Connectivity in the Commercial Internet. Michael/ Wright. Carter and Wright maintain that two-part interconnection charges could reinstate the collusive outcome. 22.): Handbook of Telecommunications Economics. in: European Journal of Law and Economics. Vol. Washington. 433-472. At a higher price. in the absence of regulation the collusive outcome is desirable to the extent that it eliminates the double marginalization that characterizes the noncooperative approach.C. the small carrier will charge a low usage price and incur an access deficit. Ingo/ Cave. Nicholas (2005): Economics of the Internet Backbone. 33. interconnection would become a profitable service for the smaller firm and a loss maker for the larger firm. Patrick. It then confirms the theory that smaller operators tend to set higher termination. pp. This contribution gives an overview on different aspects related to the economics of the Internet backbone. If termination costs differ between networks.” OPP Working Paper Series. It is concluded that despite ups and downs the Internet backbone market has been characterised by robust competition in the past. In both cases. Martin (eds. the smaller provider will use a high usage price to end-users to make the dominant carrier incur an access deficit. under two-part tariffs. Julius (2005): The Effects of Regulating Mobile Termination Rates for Asymmetric Networks. This is all that matters. pp. pp. In order to avoid free-riding on other networks and to induce optimal network investment. In the case of two-part tariffs downstream. Study for the European Commission. . Holland/Elsevier. Ken/ Schultz. FCC. in: Review of Industrial Organization. Crémer. Ralf/ Haucap.

The heuristic was tested on historical routing tables. Klaus/ Hillebrand. which is neutral to traffic balance. and mutual transit arrangements in the Internet in logical and mathematical terms. This study conducted for the Bundesnetzagentur focuses on technical foundations for realizing VoIP in integrated voice and data networks. Report for the Bundesnetzagentur. while the . in: Proceedings of the IEEE Global Internet Symposium. This groundbreaking paper. Jean (1998a): Network Competition: I. pp. but also increases the number of outbound calls and thus increases payment to the rival. Using a discrete choice model with Hotelling preferences. descriptions and analysis of issues concerned with commercial traffic exchange on the Internet. Klaus-D. network architecture. Vol. The paper tends to confirm a roughly hierarchical structure for Internet peering and transit relationships overall. Gao. and entry barriers for competitors. Hackbarth. The socially optimal transport and termination charge is set below marginal cost. On-net/off-net price discrimination (where off-net calls are more costly at retail than on-net calls due to the wholesale termination fees) contributes to higher call charges. Technical and Public Policy Issues Relating to Internet Traffic Exchange. Santander. They argue that on-net and off-net charges should be equal and below marginal costs to correct for the call externality. It emphasizes the benefits of “bill and keep” – bill and keep avoids the access bottleneck problem altogether. GSM Association (2006): Inter-PLMN Backbone Guidelines 3. and how to utilise it. softer price competition. Dieter/ Scanlan./ Kulenkampff. Lowering the retail price of calls increases the number of subscribers. It goes on to develop a heuristic inference as to whether a given connection is peering or transit based primarily on the number of connections an operator has. to compensate for more intense competition between mobile networks. Anette (2002): The Economics of IP Networks – Market. Rand Journal of Economics. and termination rates in mobile networks. Bad Honnef. Annette/ Kulenkampff. The paper describes peering. Gabriele/ Metzler. Gabriele (2006): Technical Aspects of Interconnection in IP based Networks with Particular Focus on VoIP. and the emerging IPX. 29. and network interconnection and aims to ascertain the resulting economic and regulatory consequences. Laffont. and also overall revenues and profits unless retail prices are already at of above monopoly prices. the authors derive key results in a setting where network operators cannot price discriminate. 2006. Mark with Garcia. Patrick/ Tirole. however. but does not necessitate changes to the existing CPP retail pricing regime. a system that is currently deployed and supports more than 300 attached networks. especially as far as backbone services are concerned. 2000.7. an enhanced GRX that incorporates Quality of Service arrangements and a cascading billing model. 4 April 2006. transit. network dimensioning. Study for the European Commission. they identify key behaviours. the difference between termination charges and marginal costs is likely to be larger for MTM charges than for FTM charges. with its companion (1998b). Jean-Jacques/ Rey. This may be of importance inasmuch as users are not accustomed to paying for incoming calls. 1-37. and the results were verified by a major Internet backbone operator to be generally correct. Alberto/ Hackbarth. Lixin (2000): On inferring autonomous system relationships in the Internet. Harbord and Pagnozzi show that bill and keep is able to reduce incentives for inefficient price discrimination between on-net and off-net calls.Final Report: The Future of IP Interconnection 157 Elixmann. Harbord Pagnozzi (2007): This paper addresses the relationship between on-net/off-net price discrimination. The document provides guidelines and recommendations on how to establish inter-provider interconnection. provides a comprehensive analytic framework for interconnection of telephone networks. Overview and Nondiscriminatory Pricing. This study conducted for the European Commission provides data. It analyzes different features including Quality of Service (QoS). Two models are described: The GRX. An increase in access fees will tend to increase retail prices.

In a remarkably broad range of environments.itu. Jean-Jacques/ Marcus. nevertheless an antipathy of CPP customers to the idea of paying for incoming mobile telephone calls has hindered its diffusion yet. 370-390. MacDonald. The paper develops a framework for Internet backbone competition. Jean (2000): Competition in Telecommunications./ Meriluoto. Jean-Jacques/ Rey. regulators must be equipped to . Jean (2003): Internet Interconnection and the Off-Net-Cost Pricing Principle. in: Telecommunications Policy. Price discrimination can cause misallocation of resources. J. however. Laura (2005): Efficient usage and access pricing in telephone networks. Dubai. Vol. This paper pleads for bill-and-keep regimes in mobile termination rates. it may increase welfare in the case of competition among equals. Price Discrimination. in: RAND Journal of Economics. This paper expands on Marcus (2006). Laffont. 29. In the absence of direct payments between websites and consumers. an analysis which they carry forward in the companion article.sum03. the necessity of regulating termination charges. pp. and also mitigates the risk of double marginalisation. which combines the advantages of RPP while solving its disadvantages at the same time. Vol. Finally. The model extends naturally to multiple classes of service (QoS). 30. It argues about the disadvantages of the Calling Party Pays Principle (CPP) in terms of market power and consequently. The paper finds that. CPP). Stephen C. in: International Journal of Industrial Organization. Marcus. in: Rand Journal of Economics. MIT Press. 34. This paper extends the work of the (1998a) paper to consider price discrimination. 2007. the paper compares these results to those of the early groundbreaking Armstrong (1998) and LRT(1998) papers and finds that the main difference is that in the telecommunications models there is a “missing price” such that receivers do not pay for calls (i. the author calls for the application of bill-and-keep. 242-277. available at http://www. available at: http://www. are a key theme. Laffont. J. Receiving Party Pays. Patrick/ Tirole. This paper analyzes how outgoing and incoming access and call effects can be internalized by efficient pricing in telephone networks. Rey and Tirole papers. especially in the sense of different prices for on-net and off-net calls. operators set prices for their customers as if their customers’ traffic were entirely off-net. 615-623. because it leads to more intense competition for market share.org/abstracts/abstracts/2003/rje. Jean (1998b): Network Competition: II. p. Patrick/ Tirole.. Changes in the nature of market power. Jean-Jacques/ Tirole. The Receiving Party Pays Principle (RPP) is estimated more appropriate.e. This is an expanded book-length treatment of the material in various Laffont. Vol. For operators to raise each others’ costs through high access charges does not necessarily increase profits. 2000. The paper goes on to analyse the impact of the access charge on competitive strategies in an unregulated retail environment.158 Final Report: The Future of IP Interconnection optimal lease price is equal to marginal cost. including (1998a) and (1998b). 23. Scott (2007): Interconnection in an IP-based NGN Environment. the access charge allocates communication costs between websites and consumers and affects the volume of traffic. Where the largest operator in a country fails to voluntarily interconnect with at least two or three of its most significant competitors.pdf. and the risk of new forms of market power emerging.pdf.int/ITU-D/treg/Events/Seminars/GSR/GSR07/discussion_papers/JScott_ Marcus_Interconnection_IP-based. GSR Discussion Paper. Ian A. Littlechild. pp. Laffont. presented at the ITU Global Symposium for Regulators.rje. where different classes of customers are associated with different levels of cost.Laffont. the backbones will (if they can) price discriminate to reflect the difference. pp. Therefore. Scott/ Rey. Vol. They extend the results to two-part tariffs. 38-56. (2006) Mobile Termination Charges: Calling Party Pays versus.

Accounting for use poses many difficult challenges. in global perspective. The study discusses regulatory developments in the UK as BT and Ofcom establish a new relationship (Openreach) in support of NGN deployment. Using a non-cooperative bargaining framework to .005 US for fixed and mobile have resulted in rapid mobile penetration. Paul/ Mitchell. Report for the Bundesnetzagentur. and in any case never applied to the fixed network. Geneva. low retail prices and high usage. but will not necessarily command the price premium that network operators hope for. MIT Press. There is a tendency to dismiss US arrangements merely as a consequence of the use of RPP. Padmanabhan (2000): Competitive Effects of Internet Peering Policies. J.itu.S. Marcus contends that PSTN/PLMN interconnection arrangements will become unsustainable as networks evolve to NGNs.Final Report: The Future of IP Interconnection 159 take action.) The Internet Upheaval. Accounting Systems and Interconnection Regimes in the USA and the UK. background paper commissioned for the ITU New Initiatives Programme workshop on “What rules for IPenabled Next Generation Networks?” held on 23-24 March 2006 at ITU Headquarters.htm). Scott (2004): Call Termination Fees: The U.S.org/content_e/diskus/274. Marcus advocates a migration to U. Bad Honnef. RPP has practically disappeared in the US. available at: http://www. Ingo/ Compaine. Available at: ftp://ftp. wholesale arrangements are a voluntarily negotiated outcome subject to these regulatory bounds. The paper discusses US interconnection arrangements for the Internet and the PSTN and contrasts them with those of Europe. available at: http://www.bundesnetzagentur. Differentiated QoS has merit. he recommends as an interim measure retaining CPNP arrangements but moving to very low termination fees. 175-195. provided that the largest operator in a country interconnects on an IP basis with at least two or three of its most significant competitors. This reflects experience in India – termination fees of $0. outcome..1. and consequently suppress use of mobile services.-inspired Coasian interconnection arrangements for developed countries as the migration to NGN gets under way.wik. J. Scott (2006): Framework for Interconnection of IP-Based Networks. Regulation to require interconnection should not be necessary.S. but the U. The US does not regulate mobile retail arrangements (which are mostly “buckets” plans) at all. In terms of policy.S. 2006. in: Vogelsang. Marcus.int/osg/spu/ngn/documents/Papers/Marcus-060323Fin-v2. call termination and roaming arrangements comprehensible to Europeans. This paper attempts to make the U. A migration to arrangements that reflects U. in reality. Bridger/ Srinagesh. Milgrom. experience with Bill and Keep and with Internet peering is likely to be most satisfactory at such time as networks evolve to NGNs. Marcus. For developing countries.pdf.de/pub/zewdocs/div/IKT04/Paper_Marcus_Parallel_Session. particularly in light of the need for competing operators to share sensitive data.pdf . Benjamin (eds.S. pp. This paper integrates the earlier Marcus papers. and provides a comprehensive overview of established economic theory for both the PSTN/PLMN and the Internet. Paper Presented at the 4th ZEW Conference on the Economics of Information and Communication Technologies. Mannheim.S. Current arrangements in any case leave much to be desired – they tend to result in elevated prices for mobile services. Obligations of parity and symmetry and an obligation to interconnect appear to be critical to the U. Scott (2006): Interconnection in an NGN Environment. however.zew. seemingly the best of all worlds.de/media/archive/6201.pdf. Marcus advocates a move to low or zero termination rates as a means of lowering German mobile retail prices and stimulating use of the service. the paper explores the proper role of regulation in the future world of the NGN. J.and also available as WIK Discussion Paper 274 (see http://www. 2004. Current wholesale arrangements are a consequence of a lack of a regulatory obligation to compensate at a rate other than zero. Marcus. More generally. and particularly where the mobile network is not yet completely built out.

Jeffrey H. collusion would only occur under mature competition with sufficient heterogeneity between the network services. access charges should be set above termination costs. 2002. with increasing market penetration the larger ISPs gain a bargaining advantage over smaller ones because their own customers value outside communications less highly than before. pp. and Laffont and Tirole. Among these regimes Bill and . in: Journal of Economic Literature. two-part interconnection charges could reinstate the collusive outcome (Carter and Wright. This is a concise and very readable summary of the impact that network effects have on technology introduction. Examples are drawn from telephony. Abrechnungssysteme und Zusammenschaltungsregime aus ökonomischer Sicht (Accounting Systems and Interconnection Regimes from an Economic Perspective). the use of two-part tariffs at retail could lead to access charges that would sustain welfare-optimal usage charges downstream. It sheds some light on how to regulate termination charges. making calls coming from the other network profitable. the collusion incentive could be reduced if firms subsidized their subscribers for incoming calls (Laffont and Tirole 2000). in: RAND Journal of Economics. The resulting peering arrangements (and the lack thereof) are efficient. XLI. pp. Rohlfs. Rohlfs did the groundbreaking work in this area of economics many years ago. CDs. Vogelsang. in the later stage. Second. Different industries solved the problem of the "initial adoption hump" in different ways. Vol. The book deals with the question of how new technologies overcome the initial adoption hump. Network effects create an initial adoption hump that make the introduction of new products and services more difficult. It concentrates on access price regulation in telephone networks. the benefits of “bill-andkeep” regimes are highlighted. This is a totally non-technical survey of the literature. television. Carlo (2005): Investments and Network Competition. as long as there are sufficiently many core ISPs.160 Final Report: The Future of IP Interconnection analyze the negotiations between a core ISP and other ISPs. 2000) Vogelsang also treats bill and keep and its relationship to receiver benefits. 830-862. 2003. Third. He also critiques the L-R-T (1998a) and Armstrong (1998) result that reciprocal access charges may be perfect tools for collusion. MIT Press. the authors hypothesize that in early stages of the Internet network size did not convey a major bargaining advantage so that bill-and-keep arrangements would be likely outcomes independent of relative sizes. First. in order to create investment incentives. This paper analyzes the impact of two-way access charges on the incentives to invest in networks with different levels of quality. This report for the BNetzA concerns potential interconnection regimes in an NGN environment. On the contrary. However. Ingo (2003): Price Regulation of Access to Telecommunications Networks. Due to he problems associated with setting the right level of access costs. Its results suggest that private negotiations over reciprocal access charges would not be efficient as firms prefer to set access charges above termination charges. Several end-regimes are evaluated according to a set of economic criteria that were provided by the BNetzA. and many other fields. and finds it reassuring that several features counterbalance the collusive effects. superior coverage or discrimination between on-net and off-net calls can induce the larger network operator to use interconnection for a price squeeze. In contrast to the other two extensive surveys (Armstrong. that is. Fourth. In contrast. 446-467. The intuition here is that firms would want to do that if access charges exceeded marginal costs. something worth further research. Vogelsang. Valletti. High substitutability would prevent sustainable collusive equilibria. Ingo (2006). (2003): Bandwagon Effects in High Technology Industries. 2003). personal computers. Tommaso/ Cambini. how they can be successfully introduced when they have to compete with well established technological offerings. Vol. 36. VCRs. so that regulatory intervention could be required to attain an equilibrium.

Some of these are compatible with Bill and Keep.Final Report: The Future of IP Interconnection 161 Keep in combination with an agreement about suitable interconnection points emerged as the most appropriate followed by capacity-based interconnection charges. Complications also arise from a number of quality-related issues that the author subsumes under the QoS heading. Since the current interconnection regime in the PSTN uses element-based charges for a prespecified set of interconnection points. moving toward zero. such as guarantees on speed and failure rates of messages. The author suggests a gradual move towards Bill and Keep through a sequence of reductions in the element-based interconnection charges. if any. while transit agreements through core networks should be based entirely on voluntary agreements. once NGNs are built. as long as agreements about standards can be worked out between interconnecting networks. . The author maintains that. migration problems arise that may influence the choice of the best end-regime. regulation. Others. should be restricted to access networks. are hard to implement across networks under any charging regime.

Bill and Keep: agreements to interconnect and to exchange traffic without payment. CPNP (Calling Party’s Network Pays): an interconnection regime where the network of the party who placed the call (the originating network) makes a payment to the network of the party that received the call (the terminating network). AS (Autonomous System): An independently managed IP-based network with its own IP routing policy. ARPU: Average Revenue per User. C CBC (Capacity Based Charging): A wholesale pricing regime reflecting the maximum capacity required. . typically expressed in bits per second. a system sometimes referred to as Sender Keeps All. BGP4 (Border Gateway Protocol v4): the inter-domain routing protocol used by the Internet. CPE (Customer Premises Equipment): A terminal device in a computer network.162 Final Report: The Future of IP Interconnection Annex 2: Glossary of Terms A Access: Access enables an operator to utilize the facilities of another operator in the furtherance of its own business and in the service of its own customers. The software running on the customer’s Personal Computer (PC) (often just a web browser) might be the client of software running on a server platform of the service provider. CDMA (Code Division Multiple Access): A set of standards for mobile communications. a telephone network or a telephone system which are at the end-user’s premises. A single server can support a great many clients. CODEC (coder decoder): An encoding or decoding device that enables the digitization and digital transmission of analogue information (such as voice). Client-server: an asymmetric technical implementation involving to computers whose functions are not the same. B Bandwidth: The capacity of a channel to carry information. CDMA is used in the United States and a number of other countries. ASN (Autonomous System Number): a unique numeric identifier for an AS.

Final Report: The Future of IP Interconnection 163 CPP (Calling Party Pays): The most common retail payment arrangement. whereby selected packets can be marked as having application requirements other than best efforts. An ECS is a service normally provided for remuneration which consists wholly or mainly in the conveyance of signals on electronic communications networks. by radio. irrespective of the type of information conveyed. The recipient typically pays nothing. ECSP (Electronic Communications Service Provider): a provider of electronic communications service (ECS). typically by dialing a designated prefix. DNS (Domain Name System): the system of databases which associates various sorts of information with domain names in order to translate hostnames to IP addresses for Internet access. electricity cable systems. ECNP (Electronic Communications Network Provider): a provider of an Electronic Communications Network (ECN).and packet-switched. (Framework Directive. It also stores other information such as the list of mail exchange servers that accept DWDM (Dense Wave Division Multiplexing): See WDM. D DiffServ (Differentiated Services): a IP-based data communications protocol which enables hop-by-hop traffic management. where applicable. Article 2). E EBC (Element Based Charging): A wholesale pricing regime reflecting the network elements used. and cable television networks. the party that places the call pays a usage-based price for the call. In a CPP system. including Internet) and mobile terrestrial networks. An ECN is a transmission system and. including satellite networks. CPS (Carrier Pre-Selection): A set of arrangements where the end-user selects a default telephone service provider (other than the network operator that connects the end-user to the Public Telephone Network) for all calls. including telecommunications services and transmission . to the extent that they are used for the purpose of transmitting signals. CS (Carrier Selection): A set of arrangements where the end-user explicitly selects a telephone service provider (other than the network operator that connects the end-user to the Public Telephone Network) on a call-by-call basis. fixed (circuit. networks used for radio and television broadcasting. by optical or by other electromagnetic means. switching or routing equipment and other resources which permit the conveyance of signals by wire.

which can be used to map a telephone number to a ranked list of Internet services. . FTTH: Fiber-To-The-Home. FTTx: A generic acronym that could for example represent FTTB. it does not include information society services. content transmitted using electronic communications networks and services. H IETF (Internet Engineering Task Force): the protocol engineering arm of the Internet formally established by the IAB in 1986. Article 2). FTTC: Fiber-To-The-Cabinet or Fiber-To-The-Curb. or exercising editorial control over. GSM (Global Système Mobile or Groupe Speciale Mobile): A set of standards for second generation (2G) mobile communications.164 Final Report: The Future of IP Interconnection services in networks used for broadcasting. other things being equal. G Gbps (Gigabit per second): one billion bits per second. ETSI TISPAN (ETSI Telecoms & Internet converged Services & Protocols for Advanced Networks) F FTTB: Fiber-To-The-Building. reusable modules for commonly used functions. which do not consist wholly or mainly in the conveyance of signals on electronic communications networks. but exclude services providing. Elasticity: The response of demand to price. drawing on the technology of the DNS. FTTP Fiber-To-The-Premises. ENUM: A mature IETF standards-based mechanism. as defined in Article 1 of Directive 98/34/EC. FTTC. or FTTH. (Framework Directive. FTTN: Fiber-To-The-Node. based on IP and SIP protocols. IMS (IP Multimedia System or Integrated Multimedia System): a standards-based platform. An increase in prices generally leads to lower demand. that seeks to employ common.

The MDF is a termination point within the local telephone exchange. using a 128-bit address system. IPv6 (Internet Protocol. J Jitter: Variability of delay. IP forms the basis of the Internet.Final Report: The Future of IP Interconnection 165 Interconnection: Interconnection enables an operator to establish and maintain communications with the customers of another operator. IP (Internet Protocol): The Internet Protocol is a data communications standard that allows computers to communicate with one another over digital networks. L Latency: Delay. M Mbps (Megabit per second): one million bits per second. ITU (the International Telecommunications Union): a United Nations agency for information and communication technologies whose mission is to facilitate global communications. version 6): IPv6 is the emerging protocol for transmitting Internet Protocol datagrams over the Internet. K Kbps (kilobit per second): One thousand bits per second. ISP (Internet Service Provider): An ISP is a firm that enables other organizations to connect to the global Internet. version 4): IPv4 is the current protocol for transmitting Internet Protocol datagrams over the Internet. LLU (Local Loop Unbundling): the regulatory requirement mandating certain telecommunications operators to wholesale to competitors the connections from their telephone exchange's central office to the customer's premises. . Together with the TCP protocol. using a 32-bit address system. IPTV (television over IP): IPTV is the distribution of video programming (one way) by means of the Internet Protocol. MDF (Main Distribution Frame): A signal distribution frame for connecting equipment (inside plant) to cables and subscriber carrier equipment (outside plant). IPv4 (Internet Protocol.

e. Network Provider: the organization that provides the network connectivity to the Service Platforms. O optical-electrical-optical (O-E-O) conversion: Conversion of an optical signal to an electrical signal and back again. for voice telephony. regulatory authority.166 Final Report: The Future of IP Interconnection Media Gateway Control Protocol (MGCP. It was originally designed to reduce the complexity and thus to improve the performance of routers in ISP backbones. the FCC. including signalling and session management. QoS-enabled transport technologies and in which service-related functions are independent from underlying transport-related technologies. Net Neutrality or Network Neutrality: A proposed regulatory principle the seeks to limit anticompetitive discrimination by network operators and service providers. and also to support traffic engineering. . N NAPs (Network Access Point): A public peering point.” NRIC (the Network Reliability and Interoperability Council): The NRIC is an industry advisory council to the U.g. OSI Reference Model: A layered data communications protocol model. NGN (Next Generation Network): The ITU defines a Next Generation Network as “… a packet-based network able to provide services including Telecommunication Services and able to make use of multiple broadband. MPLS (Multi Protocol Label Switching): A data communications protocol developed by the Internet Engineering Task Force (IETF).S. the value of a network to its users is greater as the number of participants in the network increases. also known as H. including authentication and authorization of the user. It offers unrestricted access by users to different service providers. MoU (minute of use): A minute of use. Network Externality or Network Effect: Where network effects are present. It supports generalized mobility which will allow consistent and ubiquitous provision of services to users. NASS (Network Attachment Subsystem): provides the Network Attachment Control Functions (NACF).248 and Megaco): A standard data communications protocol for handling VoIP Media Gateways.

This is a function of the distance that the signal must travel. Article 2(c). PSTN (Public Switched Telephone Network): the network of the world's public circuitswitched telephone networks. and the speed of light in the medium employed (typically wire or fibre). and the probability of packet loss. In an IP-based environment. such as the fraction Queuing: the need for one packet of data to wait for another in order to gain access to a shared facility. but not for third parties. These delays can be analysed using a branch of mathematics known as queuing theory. as defined in the Universal Service Directive. Peering: the arrangement whereby ISPs exchange traffic for their respective customers (and for customers of their respective customers). P PATS : Publicly Available Telephone Services. Propagation delay: The time that it takes for light or electricity to reach its destination in a network.Final Report: The Future of IP Interconnection 167 OSS (Operations Support System): A system to support network operations or management. Peering is a substantially symmetric form of network interconnection. PLMN (Public Land Mobile Network): a network to provide mobile communications services to the public. PoI (points of interconnection): A point at which networks meet for purpose of interconnection. R RACS (Resource and Admission Control Subsystem): The RACS provides the Resource Attachment Control Functions (RACF). . QoS often denotes measures of delay. Peer to peer (P2P) – a system where the users typically have a symmetric relationship with one another. It could also denote other measures of service quality. variability of delay. Q QoS (Quality of Service). including resource management and admission control based on the user’s profile and the resources currently available.

” Framework Directive. modifying. typically on a basis reflecting the call duration The calling party typically also pays. or multicast sessions that include Internet telephone calls. or SCTP.168 Final Report: The Future of IP Interconnection Ramsey-Boiteux pricing: A pricing principle whereby the service provider takes the highest price mark-ups on those services that have lowest demand elasticity. . multiparty. and multimedia conferences. Article 14(2). that is to say a position of economic strength affording it the power to behave to an appreciable extent independently of competitors. SMP Significant Market Power: A firm is “… deemed to have significant market power if. It can be used to create two-party. either individually or jointly with others. see Bill and Keep SIP (Session Initiation Protocol): an application-layer data communications control protocol for creating. customers and ultimately consumers. S Street Cabinet: a cable distribution system located close to customer premises. RBOCs (Regional Bell Operating Companies): The former Bell System incumbents providing local telephone service in the U. multimedia distribution. that is. Sender Keeps All. or between service providers. RSVP is the key component of the Integrated Services Architecture (ISA). which reflects the common understanding about the level of service to be provided. RPP (Receiving Party Pays): A retail billing arrangement in which the receiving party pays for the call. Reciprocal compensation: Reciprocal compensation fees are two-way termination fees. UDP. It is widely used as a signalling protocol for Voice over IP. SLA (Service Level Agreement): a contract between a customers and his or her service provider. and terminating sessions with one or more participants. T TCP/IP Reference Model: The layered data communications protocol model used by the Internet. where high prices will have least effect in diminishing demand. it can run on TCP.S. generally applicable to local interconnection in the United States. SIP is designed to be independent of the underlying transport layer. along with H. it enjoys a position equivalent to dominance.323 and others. RSVP (Resource ReSerVation Protocol): a data communications protocol designed to reserve resources across the Internet so as to assure end-to-end QoS for applications that require such assurances.

to exchange voice traffic. Tier 1 ISP: A large. Tier 1 ISPs are richly connected to one another by peering.11a. W-Z Wi-Fi (Wireless Fidelity): Wi-Fi is an IEEE standard adopted in 1999 for short-range wireless digital connectivity. . WAN: A Wide Area Network.11g standards.11b. It is by far the most widely adopted WLAN standard and includes the 802. WDM (Wave(length) Division Multiplexing): A technology that effectively increases the capacity of fibre optic systems by using different wavelengths (colours) of laser light to carry different signals. well-connected Internet Service Providers that has no significant need for a transit provider. 802. VPN: A virtual private network. Transmission Control Protocol (TCP): A data communications protocol used to assure reliable delivery of data in an IP network. U-V VoD (Video on Demand): Video on Demand enables end-users to select and watch video content over a network.Final Report: The Future of IP Interconnection 169 Teledensity: The level of deployment and adoption of communications networks in a given geographic area. either physically or virtually. VoIP Peering: The agreement between VoIP providers to interconnect. 802. VoIP: (Voice over IP): A set of data communications protocols and technologies to enable voice to be sent over individual IP-based networks or over the Internet.