Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
FINANCIAL ISSUES 172 5 Financial Issues The evolution of the Internet has several financial dimensions. These include, in the context of this report, the changing role, targeting, stability, and level of federal funding to support the supply and use of information infrastructure, as well as the changing exposure of Internet users in the research and education communities to infrastructure expenses on the demand side. As federal infrastructure activities expand from a National Research and Education Network (NREN) to a diffuse and private-sector-dominated National Information Infrastructure (NII) program, the research and education communities present both specific policy concerns and insights into the larger problem of financing access across the population. FEDERAL AND OTHER FUNDING FOR NETWORKING TO DATE Federal financing for information infrastructure fits into a large complex of decisions: choices are made about spending on research and education versus other activities, on different kinds of research, on different kinds of education, and on elements that enter into research and education as inputs, such as networks. Of all these choices, the federal role in supporting research in general is perhaps the most established. Decision making associated with the federal research and development budget has placed a high value on both the development and use of networks to support the research community, in particular. Federal support has yielded enormous benefits: increased understanding of how networks and network-based resources can be used, enhanced efficiency and effective
FINANCIAL ISSUES 173 ness of research through easier sharing and communication, and qualitative change in the nature of research undertaken. These are hard-to-quantify benefits that must be compared to the more obvious costs. However, the difficulty of making that comparison in practice inhibits the adoption of improved information technologies.1 In the coming months direct financial support by the federal government for the Internet will begin to diminish.2 Key developments are the shift to commercial service anticipated in 1994 for what has been the NSFNET backbone and, over the next four years, the National Science Foundation (NSF)- supported regional networks (see Box 1.3 in Chapter 1). Thus we find the uncomfortable situation that those parts of the research and education community that have benefited most from NSFNET face the prospect of changing to a new mode of payment for network-based services while presumably maintaining current commitments and continuing to contribute to the expanding information infrastructure. Although the actual amount of federal support is smaller than most in the research community (in particular) believe, the NSFNET transition has prompted considerable uncertainty and anxiety. These concerns are a legacy of federal Internet support, which has created a dependence on the Internet. Meanwhile, new federal programs (such as the Telecommunications and Information Infrastructure Assistance Program of the National Telecommunications and Information Administration (NTIA)) and new industry efforts (such as the Pacific Bell plan to wire up schools in its service areas) will result in further redistribution of the resources supporting the use of information infrastructure. Pacific Bell announced in February 1994 a $100 million program for "linkage for computer communications and videoconferencing" to nearly 7,400 public K-12 schools, public libraries, and community colleges in its service area by the end of 1996.3 As a result of these developments, some communities that have found network access difficult, such as the K-12 schools and hospitals, should gain greater access to network- based information resources and services generally and advance their use of the Internet in particular. However, actual amounts of new or redirected federal funds may be relatively limited. For example, in a February 1994 speech, President Clinton affirmed his support for extending connectivity to libraries, but added that he was "afraid that the lion's share of that work will have to be done at the state and local level."4 To the extent that redistribution of federal funds helps others to join the network without overconstraining the participation of incumbent users, benefits will be experienced more broadly. (The alternative, of course, is a zero-sum situation in which early beneficiaries, such as certain segments of the research community, suffer uncompensated reductions in
FINANCIAL ISSUES 174 support so that new or inadequately supported beneficiaries can gain support.) The broader distribution of benefits is a stated goal of current NII-related initiatives. Realizing that goal requires the integration of the NREN objectives, which were built on government funding, with the mission of commercial enterprises, which presumes private funding. Funds associated with the NREN program have been allocated for three purposes: to construct and grow the infrastructure, to pay the ongoing costs of network operation, and to underwrite fundamental research and development.5 Box 5.1 presents relevant federal funding figures. The federal government has played a leadership role in the Internet environment by contributing funds for development of technologies from protocols to routers, and for construction of such components as the NSFNET backbone, mission agency networks (e.g., ESnet, NASA Science Internet), and the Federal Internet Exchanges. Network construction investments, in particular, have been small from a national perspective (tens of millions of dollars), in comparison to investments for other infrastructure elements, including telephone and cable networks, that have been privately financed (tens to hundreds of billions of dollars). The transition to commercial Internet service shifts attention from federal investment in network construction to support for network operation, maintenance, and use. See "Influence on Network Deployment" in Chapter 6. The establishment of regional networks in cooperation with state governments and industry illustrates how the federal government leveraged a relatively small investment for the benefit of a large and dispersed community. The NSF could not afford to finance broad, direct institutional connectivity to supercomputing centers, its initial research networking concern, and so it catalyzed a set of mid-level (i.e., between the intra-institutional and national levels) networks that was charged with developing other sources of support over time (see Appendix A).6 Today there are several regional networks that individually receive the equivalent of $3,000 to $6,000 per year in backbone support from NSF along with other assistance.7 Notwithstanding the broadly recognized federal role, the Internet has benefited already from substantial private investments, often through partnerships aimed at extending the infrastructure and experimenting with new services and modes of delivery. For example, IBM and MCI contributed millions of dollars to the NSFNET backbone through ANS, in which they have had equity interests.8 The Gigabit Testbed program funded by NSF and the Advanced Research Projects Agency (ARPA), credited with accelerating the industrially driven development and deployment of asynchronous transfer mode and SONET technologies, involves substantial commitments of money, time, and talent from telecommunications companies. The launch of certain state and regional
FINANCIAL ISSUES 175 networks, such as NYSERNet, and confederations of some, such as FARNET, also drew on substantial private investment. Thus, private-public partnerships have been essential to Internet growth for some timeâthe federal Internet investment has always been bounded and leveraged. The private component will inevitably rise with investments in network infrastructure and investments in the commercial provision of information resources. BOX 5.1 FEDERAL FUNDING FOR THE NREN PROGRAM Total federal funding for the NREN program was $114.4 million in FY93 and is expected to be $142.1 million in FY94, with significant amounts from this funding expected to cover the extra costs associated with making the transition from the larger NSFNET to commercial service and a separate research network (the vBNS). The budget request for FY95 is $176.8 million. These figures cover all aspects of the NREN program, including research, application development, and network operation among the National Science Foundation (NSF), the Advanced Research Projects Agency (ARPA), the U.S. Department of Energy, and the National Aeronautics and Space Administration, which account for the largest segments of the NREN budget, and other High-Performance Computing and Communications (HPCC) agencies (the National Institutes of Health, National Security Agency, National Oceanic and Atmospheric Administration, Environmental Protection Agency, National Institute of Standards and Technology, and Department of Education). The ARPA and NSF receive the largest shares of the NREN program budget; the ARPA portion of this program was $43.6 million in FY93 and $48.7 million in FY94, and the NSF portion was $40.5 million in FY93 and $47.9 million in FY94. The new Information Infrastructure Technology and Applications (IITA) component of HPCC, intended to foster government-industry- university collaboration in the development of "technologies needed to improve effective use of the NII," was budgeted at $156.0 million in FY94, with ARPA slated for the largest segment, $95.6 million. NSF has the next- largest requested segment, at $19 million. The FY95 HPCC budget request emphasizes the IITA component in the proposed increments with an 81 percent increase, resulting in a total IITA request of $281.6 million.* * Committee on Physical, Mathematical, and Engineering Sciences, Federal Coordinating Council for Science, Engineering and Technology, Office of Science and Technology Policy. 1994. High Performance Computing and Communications: Toward a National Information Infrastructure. Office of Science and Technology Policy, Washington, D.C.; and Committee on Information and Communication, National Science and Technology Council. 1994. High Performance Computing and Communications: Technology for the National Information Infrastructure. Supplement to the President's Fiscal Year 1995 Budget. Office of Science and Technology Policy, Washington, D.C.
FINANCIAL ISSUES 176 Despite the exponential growth of the Internet, commercial networks that provide Internet service have grown more slowly than some believe they might have. The withdrawal of NSFNET backbone support was motivated by arguments that such federal funding constituted a market-distorting subsidy, inhibiting the entry of competitive providers.9 Nevertheless there are today at least three-dozen commercial providers of Internet interconnection service. These range from Internet-access specialists, such as AlterNet, ANS CO+RE, DELPHI, PANIX, and PSI; to local or regional networks, such as Westnet, BARRnet, CERFnet, LosNettos, NEARnet, NYSERNet, Sesquinet, and SURAnet; to full-service telecommunications carriers, such as AT&T, MCI, and Sprint. Entry into the Internet interconnection market indicates that commercial service without subsidy appears viable even if, as is often the case in new markets, it takes time for service to be profitable. Inferences from Internet-related business successes should be drawn with caution. Successful delivery of Internet service to the research and education communities is not necessarily an indicator that the same service will succeed commercially, because the research and education communities are not a microcosm of society. They are unusual in that their use of the Internet has been subsidized, their priorities and values regarding the kinds and conditions of network use are nonproprietary (see Chapter 4), and most Internet users are served en masse from university or industrial campuses. Competition among Internet access providers for a budget-constrained clientele suggests that there are negligible margins to support donations of services, cross-subsidies, or hypothetical new taxes or other mechanisms aimed at generating support for needy users. The planned private-sector, multibillion-dollar investments in enabling infrastructure suggest that it is easier for businesses and investors to see profit in home entertainment than in the information-sharing activities on the Internet. This situation is notable given the minimal experience with the new entertainment services touted for the information infrastructure.10 Indeed, financial analyses in the wake of the aborted Bell Atlantic-TCI merger suggest that investors may become more conservative about profits and opportunities relating to the cable industry (and even telephone companies), potentially constraining the flow of capital.11 COST OF NETWORK INFRASTRUCTURE The cost of network infrastructure is fundamental to both investment requirements and pricing options. Key cost elements include circuits, switching and other internal systems, user systems (computer and 2com
FINANCIAL ISSUES 177 munications equipment on customer premises, and associated goods and services, including training), and various high-level and application services. The relationship between cost and price is relatively well established in telephony given its long history and the influence of regulationâbut that relationship is poorly understood by many, and often seemingly indirect. The relationship is even murkier in the Internet, which is free of some of the extraordinary requirements found in telephony,12 and at best a subject of speculation in the emerging NII. Nevertheless, some examination of costs is essential to understand how different pricing options relate to economic efficiency, profitability and investment prospects, and the ability of the federal government to leverage private action in support of policy objectives. The massive programs of infrastructure development that are beginning to take shape, motivated in part by the possibility of interactive television and video on demand, could bring the benefits of Internet or Internet-like services to a much large community.13 The achievement of these benefits is not guaranteed, given that many choices must be made by multiple parties concerning architecture and deployment (see Chapter 2) and that the various enterprises might face higher costs in the short term to provide the kind of open access envisioned in Chapter 2 as desirable for the long term. Most costs of the Internet are fixed costs (that is, they are independent of usage). Fixed costs for long-distance Internet communication are dominated by the costs of transmission lines (relatively expensive) and routers (relatively cheap within NSFNET).14 The routers, which interconnect point-to-point circuits obtained from telecommunications carriers, perform packet switching. The cost of network service to any individual can be divided between the cost of network access and the cost of long-haul communication. Today's network access circuits are either dedicated transmission lines or are shared. Because the sharing that is achieved through packet switching is not common in access networks today, the costs of access are higher than they might be. By contrast, long-distance communication within the Internet uses packet switching to achieve transmission efficiencies that are not possible in the networks used for telephony (Box 5.2). Today, long-haul circuits operate at 45 Mbps, which costs, at tariff rates, about $45 per mile per month. Very approximately the total number of T3 circuit miles in the NSFNET backbone is about 16,000 miles and so might cost about $720,000 per month.15 Network operations increase this amount while academic discounts and donations reduce it. This order of magnitude suggests that the cost of long-haul communication for the several million Americans who use the Internet is on the order of
FINANCIAL ISSUES 178 BOX 5.2 THE IMPORTANCE OF SHARING A network is a shared resource. It is composed of expensive resources, such as high-speed long-distance trunks, which are too costly for even the largest users to purchase exclusively. Instead, the network is organized to share those resources among all the users in a cost-effective and productive manner. Different networks have different approaches to sharing, but the concept is always present. The idea of sharing expensive resources is a very familiar oneâmost people do not own their own airplane, hospital emergency room, or highway system. The need is occasional, and sharing is a reasonable approach. However, people also understand that this sharing can lead to limitations in access to the service: there is not always a plane going exactly where you want at the time you want to leave. And even if there is a suitable plane, there may be excess demand, in which case some potential users may be deferred or delayed. Queues for service form even at emergency rooms. In the current U.S. telephone system, there is almost always enough capacity to serve all users, but under unusual circumstances, such as Mother's Day or a natural disaster, the system can become overloaded, and some calls cannot be placed. Most consumers, even if they find this frustrating, understand that it is a reasonable compromise that produces lower overall costs. The technical term for overload is "congestion." Congestion and its control have been a major topic of study in the network research community, since proper utilization of network resources is key to cost- effective operation. Most Network Costs Are Fixed From the perspective of the network provider who actually constructs and installs the transmission facilities, most costs are fixedârelated not to the instantaneous level of usage but to long-term traffic statistics.* Rights- of-way must be obtained, workers must be paid to install and maintain facilities, and so on. All these costs are unchanged whether the links are loaded or idle. This situation implies that the network provider must manage the usage of its expensive resources, because both congestion and underutilization are undesirable. Congestion leads to user dissatisfaction, and underutilization leads to reduced revenues. So long as the network remains uncongested, the greater the number of users it supports (or, alternatively the more traffic each user offers), the greater the potential for revenue and for cost-recovery and profit. Again, the comparison with the airline industry is informative. Airlines must pay for planes and operating costs whether they fly full or empty. So they employ a variety of schemes to sell every seat. Occasionally, they overbook a flight (which is the equivalent of congestion), and to deal with that problem they offer rewards to passengers who will voluntarily defer their travel to a less crowded time.
FINANCIAL ISSUES 179 Network Traffic Is Bursty In digital packet networks such as the Internet, resources are allocated to users on a millisecond basis, whenever the computer has a burst of data to send. Measurements show that network traffic generated by computers is extremely "bursty." (That is, most of the time a computer has nothing to send, but every so often, the computer transmits a short burst of data.) This makes sharing a high-speed transmission link among many users very desirable, since each burst from a user can be sent at the link speed, so that it reaches its destination with the least delay. Not only is the offered load bursty, but it also comes in all sizes. A user doing a remote login to a host is probably generating a peak data rate of a few hundred bits per second. A video stream might be 10,000 times bigger, at a few megabits per second. And a fast bulk data transfer between today's high- performance workstations might be 100 times faster than that, at a few hundred megabits per second peak rate. This lack of a suitable model of data traffic makes the analysis of the statistics of sharing rather complex. * The number of trunks installed and the number of operators employed at any time are determined by a planning process that itself is based on measured traffic. a few dollars per person per year. However, the bigger cost barrier lies in the access circuits, rather than in the long-haul portions of the network. The committee's vision of an NII characterized by an Open Data Network (ODN) architecture and including both entertainment-telephonecable TV and general data communications services implies a need for two-way service that is comparable in speed to that needed for compressed videoâon the order of 2 to 4 Mbps, based on current industry discussions.16 At present most people manage with much more modest speeds, such as those accommodated with dial- up modems (typically 9.6 or 14.4 kbps) over voice-grade lines, because these are compatible with existing telephone plant; those homes with integrated services digital network access can communicate at 128 kbps (or twice that, after compression),17 and reasonably priced Ethernet service over cable has been introduced in portions of California and Massachusetts. But as experience with the Internet demonstrates, networked file systems quickly lead to lengthy data transfers and the need for higher transfer rates. In the near future interactive graphics and multimedia documents will be a common means of information exchange that demand high speed. Accordingly,
FINANCIAL ISSUES 180 the trend on university campuses is toward Internet access at 1.5 Mbps and 45 Mbps. Pressure for these data rates to and from people's homes will emerge (Box 5.3). The scale of the NII challenge is a principal reason that the federal government has chosen not to contract for a separate network but to transfer the burden of network construction and operation to commercial interests.18 The research and education communities illustrate the scalability problem even though they are only a part of it: these communities are widely distributed across the United States; they occupy 110,000 K-12 school buildings and over 3,000 higher-education campuses; and the research and educational infrastructure for those who have already left school involves some 15,500 public libraries. These facilities house approximately 47 million school children, 15 million higher-education students, and 3 million teachers and faculty.19 Moreover, most of these same individuals also work at homeâ children and older students do homework, and faculty grade work and prepare for the next day of teaching. Serving a widely dispersed community reduces the opportunities for sharing. If only a few sites in a community are connected to a network (e.g., only the schools and libraries), it will be necessary to allocate to these few users not only the cost of the local access link from each site, but also the cost of the link from the community to a more distant network access site. This argues that there will be considerable economies of scale in serving education and libraries through a larger NII whose economic justification comes from other commercial users. It would be prohibitively expensive to provide high-speed data communication service to a large number of widely dispersed homes and campuses for research and education purposes unless there could be some economy through sharing with other users of a nationwide data communications infrastructure. To these quantitative concerns must be added qualitative ones: the priorities, timing, and location for adding new facilities and capabilities to the commercial networks may not be directly aligned with the preferences or needs of the research and education communitiesâindeed, this differential was a driver of the NREN program.20 These realities suggest a risk that some members of the research and education community will, in the short or long run, not be served as well by commercial services as they are by the Internet today. The costs of enabling upgrades to the local loop are a key barrier to the rapid and widespread deployment of NII services. See Box 5.3. Clearly, service providers must cover their costs of capital and operation, which implies some system for charging network users. The price of access to a commercial Internet service today (like the price of access to a regional network) depends greatly on the distance to the nearest net
FINANCIAL ISSUES 181 work node. That may be tens or hundreds of miles because there are relatively few points of network connection. Geography is a major reason for cost differences experienced by the University of Colorado and the Massachusetts Institute of Technology (MIT). With the introduction of commercial Internet service, MIT, in Cambridge, Massachusetts, anticipates that its costs may increase on the order of $10,000 per year (assuming both a network point of presence in Boston and a bandwidth-driven access charging scheme), whereas the University of Colorado, in a more rural setting with greater distance to the nearest network node, estimates that its annual costs could rise by as much as $100,000.21 Averaged across all connections to commercial Internet services today, the average charge for access at 56 kbps is $15,000 per year and the average charge for 1.5-Mbps access is $36,000 per year. As access to high- speed services widens (scale again), the nationwide average cost for commercial Internet access BOX 5.3 ACCESS LINKS AND WHERE THE SHARING OCCURS Access links (the "last-mile" connections to customer sites) represent an important aspect of system costs. For access links to small sites such as residences, there are few opportunities for sharing the cost among several sources of traffic. This creates a tension between two objectives. If the access link is of large capacity, which is preferable (albeit costly) to permit bursts of traffic into the network, this raises the possibility that the traffic source could produce not bursts but a continuous stream of traffic at this high speed. Providers are thus motivated to impose a high charge for such a link. Alternatively, the access link can be engineered at low speeds, which protects the provider but limits the traffic source. One solution to this dilemma is to engineer a new form of access link that permits high burst rates, but that either prevents or charges for high continuous traffic flows.* Large institutions (for example universities or corporate sites) have an easier time with the economics of sharing. With a site housing a large number of users, the very bursty traffic flows from these many sources become combined into a somewhat smoother aggregate stream across the access link into the external network. The institution can thus justify purchase of a high-speed link and allocate the cost across the total pool of users. * Note that loop transmission architectures are complex and may require, for cost and/or efficiency reasons, electronic and/or optical multiplexing that is accomplished via equipment deployed within the loop (so that the fiber is no longer dark). Also, hybrid fiber-copper loop architectures may be feasible.
FINANCIAL ISSUES 182 at 56 kbps might drop from today's price of $1,250 per month to a figure that is more like $50 per month. Today's cost estimates should not be projected far into the future, since new technologies (such as the introduction of more fiber-optic facilities and such digital transport services as ATM) will be cheaper than digital services over voice networks. Both network-and information-related costs are likely to change, for such reasons as increasing competition, deregulation, changes in enabling technology, and increases in scale. Emerging plans for reconstructing the cable TV and telephone access networks will enable at least 1.5-Mbps access for each video channel; they open the door for interactive services.22 Reconstructed access networks for cable TV, for example, should be able to support Internet access and Internet-like services at much lower cost than is evident today.23 Efforts are already being made to exploit such connections using special hardware (such as new kinds of modems supporting Internet access over cable systems) and services.24 Although the above discussion has focused on line costs, another cost element, under the current Internet structure, is associated with the physical and business aspects of the interconnection of regional, commercial Internet access, and other networks. At present, interconnection is provided by the Federal Internet Exchange (serving federal backbone networks) and the Commercial Internet Exchange (serving commercial Internet access providers) locations. The recently announced interregional CoREN is intended to serve regional networks, and the new NSF plan calls for network access points (NAPs) to interconnect various networks with the anticipated vBNS and with each other. Interconnection entails associated equipment and labor, which create costs. In the telephony environment it has also involved other costs (e.g., settlement payments covering traffic transiting the interconnection), some associated with regulation. Added cost elements in the telephony environment contribute to the observed differential in cost between sending a fax via Internet versus via the telephone network. See Box 5.4. As the shift to commercial service unfolds, there is no reliable information on how prices will change and what impact those changes will have on patterns of network use. To understand and plan for both the Internet and the NII, it would be useful to have a comprehensive study of Internet service economics, how the cost of that service can be expected to evolve in the coming years, and how it will be affected by continuing growth of the user population, the shift to multimedia communication, and the restructuring that is taking place in the telecommunications industry.25 Some preliminary efforts at relevant analyses have begun, but comprehensive and authoritative analysis would help to dispel myths and assess the cost-effectiveness of potential courses of action.
FINANCIAL ISSUES 183 BOX 5.4 INTERCONNECT CHARGES IN THE SHARING PARADIGM Issues of sharing and congestion also arise inside the network. Because the network is not implemented as one monolithic entity but is made of parts implemented by different operators, each of these entities must be separately concerned with achieving good loading of its links, avoiding congestion, and making a profit. The issue of sharing and congestion arises particularly at the point of connection between providers. At this point, the offered traffic represents the aggregation of many individual users, and thus cost-effective sharing of the link can be assumed. However, if congestion occurs, one must eventually push back on original sources. Options include pricing and technical controls on congestion; there may be other mechanisms for shaping consumer behavior as well. Since each of the network providers carrying the user's traffic must recover costs, the user payments must be shared among all the relevant network providers. One way for this to be accomplished is for each provider who carries traffic for a user to bill that user separately. However, this approach presents a serious technical challenge for connectionless networks, such as those that use the Internet protocols, since the network protocols of today (such as the Internet protocols) do not have the necessary information in the data packets to permit trustworthy accounting, and accounting at the packet level may require mechanisms of considerable cost and complexity. The alternative in use today among the interconnected providers is flat-rate payments. This approach is consistent with the observation that most costs are independent of level of usage, and it provides to both parties the assurance of known payment levels. PAYING THE PRICE Whereas the committee's vision of an open NII suggests ready access to the network by individuals wherever they may be, beginning with their homes and places of work, the goals of research and educational networking have centered on access by an institution (e.g., a school or a laboratory), which in turn provides access to individuals within the institution. Having recognized this, the NSF Connections program (and even the NTIA Telecommunications and Information Infrastructure Assistance Program and other elements of the administration's NII proposals) focuses on achieving some kind of an institutional connection as a first step; similarly, NSF's Stephen Wolff has spoken26 of a three-pronged strategy for community access: public-access terminals in a school, a public library, and a county extension office.
FINANCIAL ISSUES 184 The apparent institutional focus arises not because it is sufficient, but because it is more affordable and practical than attempting immediately to connect all relevant individuals within publicly supported education, research, library, and health-related institutions.27 Also, institutions typically provide the needed expertise for support; a substantial but often unrecognized cost relates to training and ongoing support for users.28 However, it is important to recognize that institutional access presents a different concept of national scale. It also is associated with certain kinds of gaps in support, such as the lone researcher at a small college who might not be able to get access (whereas new investigators or investigators in new fields at connected institutions might). That is, all institutions are not alike: some institutions are able to afford good access and some cannot afford any.29 Moreover, those with difficulty affording access, per se, have difficulty affording those complementary goods and services that allow them to realize the benefits of access. These range from adequate local infrastructure, to user training and support, and to access to certain information resources and services available over the network. The Pacific Bell proposal to provide connectivity for public schools and libraries was notable for its provisions to "field dedicated resource teams" to assist in the use of the capabilities being provided as well as to collaborate with universities and colleges to enhance the treatment of educational technology in the teaching curriculum. Today, most individual users of the Internet gain access through the institutions that employ or educate them and do not see the costs or charges associated with their individual use of the network. Institutions for research and education do pay for backbone access: they typically pay access charges, often through a regional network, and they have had free use of the backbone network, which supports long-distance communication. By contrast, commercial users of the Internet (e.g., corporations) pay both for access and long-distance communication (since the NSF's acceptable use policy has restricted use of the NSFNET backbone). Many institutions, such as high schools and public libraries, have only indirect access to the Internet (through a time-shared machine attached to the Internet and operated by a service provider, usually accessed at low speed, e.g., 9.6 kbps or less). These users typically pay for dial-up access (telephone service) and pay a fee that is a function of the time spent using the network.30 Depending on location and usage, such connections can be very expensive. For example, a pilot project for connecting rural libraries in New York state to the Internet found that several participating libraries were paying $175 to $350 a month in long-distance telephone charges because there was no point of presence in their local area, a figure that could increase with usage.31 Economic analysis argues that individuals must encounter accurate
FINANCIAL ISSUES 185 costs in order to make appropriate choices or trade-offs among alternatives and in order to be as effective as possible in their use of whatever level of funding is available. If alternatives appear to be free, individuals will choose those that appear the best to them, regardless of cost; a free alternative is always consumed to excess. If A and B are both free, A and B are consumed to excess. If A is free and B has an appropriate cost-based charge, A is overconsumed and B is either underconsumed or consumed at a proper level. Thus, it can be hard to assess what choices would otherwise be made in an environment where cost or price information is distorted. Among the choices to be made within the research and education communities are efficient trade-offs among the various inputs used in producing research, education, or library services, including those that complement the use of networks (e.g., desktop systems and other resources for gaining access) and those that substitute for networking (e.g., other means of communication and gaining access to information). The provision of access to the Internet without recouping full costs from users (institutional or individual) laid a foundation for two phenomena. One is the extraordinary growth in networking and applications that excites people about the Internet. The other is the growth in a sense of entitlement to low-cost (even no-cost) access within the research and education communities. These outcomes stem from the focus on the Internet as a research-related enterprise; "free" access to ARPANET and its successors was granted to the institutional communities not only to facilitate the conduct of research, in part by broadening access to such shared resources as federally funded supercomputer centers, but also because of the research value of seeing how use would develop. This access subsidy was apparently provided without making compensating changes either in the overall levels of institutional support or in the allocations of that funding among inputs for which network access and network resources were substitutes or complements. This situation implies both that there was a net increase in the effective level of funding to the institutional community and that allocation decisions among inputs have been distorted. For example, more network resources and their complements are being used than would be used (even at the same effective overall funding levels) if direct payments for such resources were required. Thus, the Internet experience with respect to cost supports conflicting interpretations: as an experiment, it suggests a potential for substantial increases in the amount of networking, which is believed to generate increases in the quality and value of network-based activity, but the basis for that expectation lies in what economists might view as excessive use of networking to date, inasmuch as pricing to users was below cost. How
FINANCIAL ISSUES 186 ever, as in other network settings, the Internet experience also points to the benefit of initial subsidies to jump-start use in situations where the value of consuming a product (in this case, networking) increases as a function of the participation of other consumers. Imminent Short-term Increases With the move to commercial service, institutions are expected to absorb the costs previously covered by the NSFNET backbone payments and those covered by a declining schedule of support payments for the regional networks. Together, these NSF payments amount to about $30 million;32 divided among connected universities, the increase in cost is expected to average a few thousands of dollars. Although absorbing this increase may cause some hardship in the short term due to transitional budget-juggling33 or for institutions for which service is inherently costly (e.g., those far from network access points), the order of magnitude is sufficiently low to suggest that there will be little hardship on average for institutions already connected to the Internet. In addition, NSF has discussed recouping funds from the scheduled reductions in support for regional networks and making them available on a competitive basis to institutions seeking to effect high-bandwidth connections. Thus, the change in NSFNET support is causing more apprehension than circumstances appear to warrant. RECOMMENDATION: Transitional Support The committee recommends that temporary subsidies of education and research institutions be considered in cases where the commercialization of the Internet generates exceptional funding distortions. The last decade has seen a powerful transformation of the activities of higher education and research in those segments where networking has become integral. Side effects of the shift away from a federally funded NSFNET backbone may include temporary disruption and financial hardship for some members of these communities, although overall the impact is expected to be limited. Costs of Local Infrastructure and Access to Services There are two other elements of concern to Internet users: local infrastructure, and the costs of accessing services once network access is
FINANCIAL ISSUES 187 achieved. Although financial support for the NSFNET backbone and the regional networks constitutes a significant amount of money, it is dwarfed by other elements of the cost of using the Internet. Wide area network service typically does not include installation, operation, and maintenance of the local infrastructure on customer premises. Local area networks (LANs), workstations and other desktop equipment, communications and interface software, and the staff that manage or support these systems cost far more than wide area network access and use.34 For example, one industry survey suggested that the typical corporation's average five-year cost of owning a personal computer is $40,000, reflecting substantial installation and learning-related costsâwhich have risen over time with the complexity of the technology.35 This situation is recognized in research as part of a larger, chronic problem in attaining and upgrading infrastructure (facilities, hardware, software, and support). Local infrastructure elements must usually be paid for out of funds that are distinct from those that pay for wide area communication obtained from a commercial provider. The cost of the local infrastructure is of particular concern to people building up the capability to use the Internet for the first time. For example, it is particularly acute for the K-12 education community, which has lacked adequate electrical power and telephone access, let alone access to computing and communication-related equipment, software, and technical support. A pilot project to connect rural public libraries to the Internet involved an investment of about $3,000 for a single basic package of hardware and software per site.36 The absence of targeted support for local infrastructure means that observed spending reflects trade-offs and choices made among a wide set of alternatives, given the available budgets. However, those trade-offs and choices may be uninformed when decision makers have not experienced network access. Experimentation, yielding insights into both the potential benefits and the costs of information infrastructure, may help people to learn how to lower costs. Local infrastructure is essential because many of the most visible networking services involve capabilities provided outside the backbone network and often go right to the end user's computer.37 Capabilities such as multimedia electronic mail and distributed database retrieval place a very heavy reliance on the end user's systems; the networks are relegated to pathways for the flow of bits, while much of the action (and investment) takes place outside the network. Thus, a critical infrastructure challenge and cost is in developing the intrasite infrastructure. In California, for example, that challenge pertains to 136 higher- education campuses and approximately 7,500 school sites and 800 libraries; a rough estimate is that the cost will exceed $2 billion using today's technologies and costs. Assuming these costs can be met, the availability of small-
FINANCIAL ISSUES 188 scale, personal computer-communications systems is what enables individuals to introduce new information resources and services, one of the major benefits associated with the Internet and the NREN program. It is important to note that the cost of network access is only a piece of the cost of information infrastructure accessâthere are additional costs associated with access to and use of information resources, costs that may be separate from or bundled in with network access. Database vendors, for example, provide access to their offerings at prices that may include a monthly subscription fee plus connect time and document or page charges or, increasingly, may be priced on a flat-fee (per month or other subscription-period) basis.38 Because of the economies of scale in the production of information, varying prices charged to different classes of usersâincluding charging lower prices for some who otherwise could not purchase the informationâmay be desirable for optimizing the production of information products.39 Although the price charged reflects many considerations (ranging from ownership of intellectual property to the costs of assembling and packaging the information for customers), the actual costs of getting into and extracting information from a database are very, very low. Most of the marginal costs for an individual user are associated with physical access (including, in addition to communications links, the software that adds a user's address to the network for billing and connecting purposes). Not all information resources require additional payments. In the Internet environment, there are many information resources offered without charge by individuals, government agencies, and a variety of organizationsâdata, research results, "shareware," "freeware," certain government information (increasingly becoming available over the Internet), and so on. Even in the commercial environment, advertiser-supported services allow for some resources to be available without additional charges. It is easier to gauge changes at the institutional than the individual level, although both will affect the level and kinds of use of networking in the research and education communities. Within institutions, questions remain about how increases in costs will be absorbed, what administrative and management changes will be made in response to an increase in costs, and the extent to which individuals who are disproportionately large users of network services may confront sudden large bills for their use. In this area, the central questions are, (1) Should individual users face charges based on their level of use? and, (2) Is there a preferred approach to pricing for infrastructure access and use? The second question, in turn, has two parts: (1) How should individual use charges be covered? and, (2) Is there a preferred mechanism for delivering a subsidy, if any?
FINANCIAL ISSUES 189 Usage-based Pricing Given the "bursty" nature of data communications traffic (see Box 5.2) and the wide variation in demand for capacity over time and in terms of bandwidth, any network that attempts to realize effective sharing must include sophisticated mechanisms to shape user behavior . The rate at which traffic is offered must be regulated to generate reasonable link loadings without serious congestion. These controls can take several forms, including pricing and technical mechanisms.40 Another argument for user charging is the expectation that, with finite resources, achieving a truly national information infrastructure will require as broad a base for cost recovery as possible. If the NII is really going to benefit a wide range of people, revenues should be derived at least partially from the individuals using the services. The High-Performance Computing Act of 1991 contemplated that the NREN could "have accounting mechanisms which allow users or groups of users to be charged â¦ where appropriate and technically feasible, for their usage of the Network."41 See Box 5.5 for a list of key cost elements in the Internet context. As noted above, economic analysis argues for usage-sensitive charging where costs are usage-related. Such charges both enable users to choose appropriately among alternative purchases and provide a check on excessive network use. If unlimited network use is available for a flat fee, there may be little incentive for restraint; in the network context, the result is congestion, as on an overused freeway, which imposes costs on other users. Specifically, those who overuse the network may reap all of the benefit but suffer only a fraction of the cost imposed on the network in the form of congestion.42 Congestion has not been a major problem on the Internet recently, both because of increases in capacity intended to accommodate growth (and overcome past congestion problems) and because most applications have been tolerant of the congestion control mechanisms in the network, which delay traffic in response to heavy demand or load. Since most Internet costs are relatively independent of level of use, in the absence of congestion the incremental cost of extra traffic is essentially zero.43 Internet congestion appears to depend on the time of use relative to periods of peak load. This situation is similar to other network contexts, where, for example, variation in pricing for peak and off-peak use can help to balance network load and accommodate users with fewer resources. However, some applications (such as nontime-sensitive file transfers) are more suitable for off- peak timing than others (e.g., real-time communication, some e-mail). The Internet has evolved, perhaps by accident, with a payment model based primarily on a fixed price for access and a service model that
FINANCIAL ISSUES 190 shares the available resources equally among all those who at one moment want to use the network. In times of low traffic, the amount of service that any person gets under this regime is not determined by the funds that he or she has available but by the capacity that exists to serve that user. (See discussion on best-effort versus guaranteed bandwidth service in Chapter 2.) By contrast, the telephone network charges either according to the amount of use that is made of the network, or at rates independent of usage (at least up to some volume); no matter how many users there are, it offers a constant quality of service. The Internet offers variable-quality service (and uses technical means to control congestion) for a fixed price, and the telephone network offers a fixed quality of service priced according to the amount used. BOX 5.5 MATCHING PRICES TO COSTS In general, the prices that users face should reflect the resource costs that they generate so that they can make intelligent decisions about utilization of resources. In the case of the Internet, there are several costs that might be considered: â¢ The incremental costs of sending extra packets. If the network is not congested, this is essentially zero. â¢ The social costs of delaying other users' packets when the network is congested. This is not directly a resource cost but should certainly be considered part of the social cost of a packet. Users bear this cost through delay and dropped packets and would often be willing to pay to reduce congestion. â¢ The fixed costs of providing the network infrastructure. As has been observed, this is basically the rent for the line, the cost of the routers, and the salary for the support staff. â¢ The incremental costs of connecting to the network. Each new user (or network of users) connection to the Internet involves costs for access lines and switching equipment. â¢ The cost of expanding capacity of the network. This will normally consist of adding new routers, new lines, and new staff. SOURCE: MacKie-Mason, Jeffrey K., and Hal R. Varian. 1993. "Pricing the Internet," prepared for a conference, Public Access to the Internet, John F. Kennedy School of Government, Harvard University, May 26-27, 1993. Prepared April 1993; November 26, 1993, version. This distinction will become less clear-cut and less sustainable as workstations are equipped to support audio and video communication. Internet users will probably wish for predictable quality of service as an
FINANCIAL ISSUES 191 option when they begin to use their computers as substitutes for telephones and videophones and as audio, video, and multimedia applications proliferate. Moreover, it would be more efficient to assign higher priority to scarce bandwidth resources for services with a higher value (such as real-time medical imagery versus leisure-time game images).44 Both schemes have their merits and should be available in the Internet after the switch to commercial service; the implication is that there will be correspondingly different pricing schemes for different kinds of use (e.g., very low prices for text e-mail and higher prices for real-time video). 45 It is useful to point out that best-effort service and flat-fee service are not synonymous; it is possible to have usage-based best-effort and flat-fee guaranteed service arrangements. Note also that any kind of traffic-or usage- sensitive pricing scheme presumes that detailed accounting information is available to support it. Pricing structures and their relationships to different models for delivering services constitute an area under discussion within the Internet community.46 Flat-fee Pricing Most research and education institutions are highly resource-constrained: they depend primarily on public financing, they must hold to fixed annual budgets, and, like many in industry, they have difficulty with costs that are not predictable over the course of a year. Research libraries' experiences with direct charging for commercially supplied information servicesâwhich libraries consider very expensiveâand recognition by some researchers that they generate substantial volumes of traffic lead to deep concerns about the volume and incidence of direct user charges. Consequently, research and education institutions would prefer to obtain networking and information services for a flat annual fee rather than be charged directly and variably for the traffic that they generate. Commercial providers of Internet access, like providers of telecommunications services generally, offer a range of charging schemes; variation in pricing is one dimension along which providers compete.47 There are flat-rate schemes that vary according to the speed of the access line (which is presumed to correlate with volume of communications traffic), and usage- sensitive schemes that are based on the number of messages transmitted (or other transactions), the time spent using the network, or the number of packets transmitted. There is a trend toward flat-rate schemes, in response to emerging commercial customer demand and complaints that usage-sensitive charging is too expensive and inhibits use.48 Different payment structures share risk between the user and the provider in different ways. A usage-based structure has risk for the user,
FINANCIAL ISSUES 192 in that unexpected usage (e.g., from misunderstanding or poor planning) can generate large expenses, which may not have been budgeted. At the same time, usage fees have risk for the provider, in that if the users curtail usage to control costs, the revenue to the provider will drop. Flat fees or access-based rates have the mutual advantage that the cost to the user and the revenues to the provider are known in advance. In exchange, flat fees offer the risk of congestion, if the users are not prevented from offering excess traffic. With flat-rate schemes, the provider assumes more risk (associated with variation in usage) than with usage- based schemes, but the impact depends on how much cost varies with useâand for much use there is little such variation. Also, low-volume users effectively subsidize high-volume users when both groups pay the same fee. One issue as these developments unfold will be how costs vary by kind of traffic. Technically, a bit is a bit, as is observed in Chapter 2, but voice and video bits are priced very differently today. Voice-level charges discourage video traffic (which requires far more bits), whereas video-level pricing would allow voice to be carried essentially for free. These differentials contribute to carrier concerns about proposals to unbundle the basic transmission facilities from some form of basic service. Given observed market conditions, the committee assumes that a variety of pricing schemes will be available to consumers of Internet service.49 It recognizes that flexibility in the way that Internet service is priced would allow those that make light use of the network to avoid the extra cost implicit in a flat- rate price that is also designed to accommodate heavy users. Flexible pricing also could permit a situation in which those with fewer resources receive lower- quality service or service with fewer guarantees as to delay or bandwidth, while those with more might achieve more quality or service guarantees; with flexible pricing there might be a series of service packages offered at different prices. A set of service packages with different flat fees and corresponding different levels of service covered by the fee would be desirable. This would allow users to be differentiated according to need, so that a person who wants to send video will pay more than a person who sends only electronic (text) mail. The concept is consistent with the two-tier service recommendation of the 1988 Computer Science and Telecommunications Board (CSTB) report, Toward a National Research Network;50 a proliferation of users, capabilities, and services (both current and anticipated) suggest a need for several pricing packages, not just two. Network providers can use pricing and other means to balance load and to shift usage to off-peak periods.51 In addition to congestion-related variation in pricing based on time of use, suggested above, another timing issue relates to making long-term versus spontaneous arrangements to use above-average amounts of network capacity. Users who antici
FINANCIAL ISSUES 193 pate only occasional demand for very high speed service can be encouraged to make an explicit reservation in advance, as an alternative to paying for the option of sending that traffic spontaneously. It is necessary to determine how to handle a situation in which someone generally has low-level needs but, for example, once or twice a year wants to do a video conference.52 A flat fee for network access offers other advantages relating to the larger goal of developing the NII. Experience with data networks is still developing, and people are uncertain about how much network capacity is consumed when a particular program or application is executed. Flatfee charging removes the tension that is otherwise present in these circumstances, and it frees the user to experiment with new capabilities without fear of bankruptcy. Moreover, as noted above, although costs of new applications involving audio and video will be relatively usage-sensitive, dominant applications today (e-mail, file transfer) typically are not. In contemplating pricing options, there is an assumption that growth in the volume of users will result in lower average prices, given that there are economies of scale. Another assumption is that an increasing number of service providers creates competition that will also place a check on prices. Further, pricing schemes may motivate new protocol developments that can result in lower costs. All may take time to unfold, and neither assumption is inconsistent with some degree of continued government support for research and education users over the near term. As a result, the ideal pricing structure is likely to change over time. The argument for flat-fee charging, in the near term, to stimulate experimentation and use, extends beyond the research and education communities to business and the general public. Usage-sensitive pricing might be preferable in the long run, when network-based applications are as mature and familiar as voice telephony, photocopying, postal service, and other information-related transactions for which usage-sensitive charging already exists.53 Covering User Charges (Subsidies and Mechanisms) Although the level of financial support is relatively small, symbolically the withdrawal of NSF support for the NSFNET backbone and eventually the regional networks signals a need for users to acknowledge and cover the costs of their use of network-based resources. In the public financing context of research and education, many fear, this situation may constrain the use of networking and network-based information resources inasmuch as budgeting is zero-sum in the short term. However, other current cost factors may change (for example, changes in state reg
FINANCIAL ISSUES 194 ulations may lower costs to research and educational users), and spending on network-based resources may result in less demand for other resources (such as books, as has been seen in the library environment). Eventually, use of network-based resources in research and education may be budgeted like use of other communications and information resources. This has already happened in some instances, and it is typical of for-profit enterprises. On the other hand, a number of mechanisms are available to support networking and access to networked information resources should such expenditures be targeted as a matter of policyâwhich would be consistent with the overall policy decision to promote an NII. The drawback to subsidizing infrastructure-related expenses specifically is that there are many information products involved in education and research, and subsidizing one subset is done at the expense of others that may be at least as useful. Many academics receive federally funded research grants awarded competitively based on proposals. A grant proposal typically includes line items for cost components such as travel, telephone use, photocopying, and computing. There is also provision for institutional overhead. In the future, network or information infrastructure access will probably show up either as an additional line item or as a component of the institutional overhead. As it is, today mission agency (e.g., DOE, NASA) funding is generally not directed to network users as such (except in the case of backbone support), but to research principal investigators who must decide on the prioritization of network needs versus other programmatic needs, given some constrained funding level.54 The advantage of including network and information access along with other cost items is that it allows the individual researcher to choose how best to allocate resources and to make choices among networks, telephones, and other alternatives. Practical arguments are often mounted against a grant-based approach. For example, if networking is treated as a line item, there is some concern that the resources available to the recipient will be reduced to cover institutional overhead costs, as happens with other kinds of expenditures. Another concern with individual project recovery is the administrative burden of tracking yet another cost item. These are among the arguments that were advanced in opposition to the suggestion made in the 1988 CSTB report to provide network- service vouchers to individual researchers to be used in covering network- related costs. Also, as discussed in "Equity" below, a research-grant vehicle could serve only that portion of the research and education communities that receives such grants.55 It is important to distinguish arguments raised against a specific kind of mechanism (e.g., a voucher for an individual researcher) from arguments concerning a generic type of mechanism. As MacKie-Mason and
FINANCIAL ISSUES 195 Varian point out, the concern about Internet access by "poor" users relates to the distribution of wealth, not pricing. Vouchers or lump-sum grants are often recommended by economists as a relatively efficient mechanism for redistributing resources.56 Vouchers effectively provide a new currency, in this context one useful only for network-related services. Given the administrative problems in routing funds to individuals, it may be more convenient to target funds to service providers rather than to end users directly. This approach seems relatively straightforward in the case of network service providers, but raises additional implementation considerations if "service provider" is expanded to include information service provider.57 That convenience has been implicit in federal procurement of network facilities and services through the NREN program to date. One possibility tied more closely to infrastructure use is a system of vouchers that are directed to service providersâa much smaller group than the end-user populationâbased on the number of subsidy-eligible users they serve.58 Such a system would place the administrative burden on the service provider, while allowing users the freedom to choose a service provider and thereby preserving competition among providers for customers. A limitation of a provider-based voucher system is that it covers only a portion of the relevant costs, the costs of the network-based service. Local costs for equipment, software, and support as well as costs for information resources would still have to be covered through conventional budgeting, including grant provisions and other sources of funding. Deriving Specific Funds Compounding the distortions that may arise from targeting funds for use of information infrastructure are problems in generating those funds. After considering the pros and cons of specific options, the committee decided against recommending specific options for generating funds that explicitly or implicitly amounted to taxation. For example, taxing communications providers (including increasing existing taxes) could yield revenues to defray expenses incurred by research and education users of the Internet, but at the likely cost of slowing investment, raising prices, and triggering other side effects common to excise taxes. If there are to be subsidies for research and education users, consistent with the NREN program and also with the connectivity goals advanced for the broader NII, funding out of general revenues would have the benefit of encouraging more explicit balancing of claims for limited federal funds. (The committee recognizes that there are ongoing discussions in government and industry on possibilities for a broad-based pool to support universal access for low-income individuals.) General reve
FINANCIAL ISSUES 196 nues and tax incentives also allow for a broader spreading of costs commensurate with the broad distribution of benefits anticipated for the NII. Equity A central issue in the move toward greater user responsibility for infrastructure expenses is the incidence of costs. (See "Equitable Access" discussion in Chapter 4.) There is concern, under the new arrangements for Internet service, that what users get may depend on how their work is funded. The NREN program illustrates one divide-and-conquer approach: the NSF intends to provide direct support for very high bandwidth scientific research. For this it has commissioned construction of a small network, the vBNS. Other Internet users will receive backbone network service from commercial providers. Clearly, NSF recognizes that some needs are easier to meet "affordably" through the marketplace than others. It implies that as a matter of public policy, some big users will receive substantial direct public support for their networking. The rationale for this arrangement reflects the nature and value of the research program. To the extent that a funding agency has redirected funds to cover the cost of Internet use, grantees may see no substantial change as a result of the switch to a commercial Internet and user charges. If you do research and your funder thinks that you need high-powered networking, your grant may provide for vBNS access privileges, you may get more resources for networking, you may get access to ESnet or NASA Science Internet. If your state government or institution wants to play in the information age, you may get basic connectivity, perhaps more. But if you don't satisfy a funder's test and if you don't live or work in the right place, you will have to finance access and services yourself. Individuals in the research and education communities routinely confront differences in financial support, both by field and by institution, and in what is considered affordable. Even telephony is regarded as expensive by some school administrations. High-energy physicists and earth scientists have relatively generous support for networking, and for computing, whereas historians are more likely to have none. A computer scientist in one of the leading university departments would expect to have a leading-edge workstation, whereas a computer scientist in a four-year college may at best have access to a lesser machine, and those studying natural languages may have access to the most basic of personal computers. Within the research community, some disciplines have a relatively healthy level of grant money that could be used to support networking needs, while other disciplines are more resource-constrained. Isolated researchers in rural small colleges face greater difficulty gaining access to networks and using them to collaborate than others better situ
FINANCIAL ISSUES 197 ated. Budgeting for K-12 education, small colleges, and public libraries tends to be even more constrained than for research. Recognition of these disparities fueled the growth and broadening of the NREN program from its roots in supporting supercomputer center access for an elite group of researchers. The discrepancies among specific disciplines or institutions may be seen as a reflection of higher-level budget decisions about where limited federal funds are best spent, allocations that result from the political process of determining where federal and state funds are spent among research, education, and other alternatives and within the categories of research and education. In that context, there are winners and losers, richer and poorer groups. The difficulty faced by K-12 schools, small colleges, and public libraries in becoming connected flies in the face of a fundamental premise of the U.S. public educational system: there should be equal opportunity for all studentsâin large schools and small, in rural areas and in the cities, regardless of income. Observed lack of such equality affects access to books, teacher salaries, athletic programs, class sizes, and so on, which compete with Internet access for scarce resources. Many within the research and education communities believe that the research and education enterprise at large benefits from the broadest possible involvement in information infrastructure. Achieving that broad involvement will require some means of addressing the financing needs of those unable to afford to participate, recognizing that such subsidized participation may be limited in scope or volume. The payoff may be neither direct nor immediate, but it may involve a lowering of costs and federal financing needs over the long term, as suggested by telecommunications economist Roger Noll: An educational system that produces a declining fraction of people who think reasonably clearly about decisions under uncertainty and about making inferences from information is fundamentally inconsistent with an information technology sector that offers increasing rewards that are available only to the educated. Of course, federal policies are always in some loose sense competitive substitutes, but in this case the complementarities are especially strong. A public policy that emphasizes enhancing the network and high-end terminal equipment largely for business purposes, but that is not accompanied by a program to enhance public education, and in particular to get computers into the schools and to create data bases and software for use by ordinary citizens, is unbalanced at best. The social value of new information technology is maximized if the maximal number of people can use it, and the act of maximizing the number of informed users also serves to minimize the extent to which redistribution of wealth will motivate the adoption of the technology. Thus better education in how to use information is an essential component of rational public policy toward the information sector.59
FINANCIAL ISSUES 198 NOTES 1. According to Roger Noll, "[An] important characteristic of information is that it can be used to make better use of other valuable resources. Information that saves time or leads to better decisions increases the opportunities available to a user, thereby reducing costs in the sense the term is used in economics. Thus, a more expensive computer or a telephone network that can transmit more information per unit of time may reduce costs. But in a large organization, an office head contemplating the purchase of a new computer or telephone system may see only a budget constraint, and not the implicit value of the increased productivity of office employees that these technologies allow. And, a state public utilities commissioner may be held accountable for only the price increase that must accompany an enhancement to the network if the telephone company is to remain solvent, not the reduction in total costs that the enhancement can bring forth" (p. 29). Noll, Roger G. 1993. "The Economics of information: A User's Guide." Pp. 25-52 in The Knowledge Economy: The Nature of Information in the 21st Century. Institute for Information Studies, Nashville, Tenn. 2. Federal support for elements of today's Internet has been reduced in the past, from decommissioning of the ARPANET to diminishing support for Internet engineering and standards-setting activities, a trend that led to the establishment of the independent Internet Society. 3. The Pacific Bell announcement was linked to a state government challenge to link and equip every classroom in California for "full-speed superhighway access by the year 2000." To meet that objective, the Pacific Bell announcement anticipates not only basic data and video connectivity, but also access to video-on-demand and other forms of interactive multimedia for public schools and libraries as the company deploys its interactive broad-band network. Assuming the proposal is approved by the California Public Utilities Commission, Pacific Bell would wire targeted locations within each institution, install service (four ISDN lines) for free, and waive usage charges for one year after installation. It also intends to work with the Public Utilities Commission to develop a special educational access rate. "Pacific Bell to Link Public Schools and Libraries to Communications Super-highway," February 14, 1994, press release distributed electronically. 4. White House Office of the Press Secretary. 1994. "Remarks by the President in Satellite Meeting with California Newspaper Publishers," February 12, electronic distribution. 5. As noted in Toward a National Research Network, advanced NREN (the original "phase III") technology requires research and innovation. Computer Science and Technology Board (CSTB), National Research Council, 1988. Toward a National Research Network. National Academy Press, Washington, D.C. (CSTB became the Computer Science and Telecommunications Board in 1990.) 6. Mandelbaum and Mandelbaum describe how both the regional networks as they are currently organized and the larger threet-tier model for NSFNET were "not an a priori construct." Mandelbaum, Richard, and Paulette A. Mandelbaum. 1992. "The Strategic Future of the Mid-Level Networks." Pp. 59-118 in Building Information Infrastructure. Harvard University Press, Cambridge, Mass. 7. Briefing to the committee, June 1993, Stephen Wolff, National Science Foundation. 8. The Merit proposal to NSF included provisions for $5 million from the State of Mich- igan for facilities and personnel, approximately $6 million from MCI in reduced communication charges, and $10 million from IBM in equipment, installation, maintenance, and operation. In September 1990, Merit, IBM, and MCI formed Advanced Networks and Services (ANS) Inc., a nonprofit corporation that was assigned Merit's responsibilities under the Cooperative Agreement with NSF. See Office of Inspector General, National Science Foundation. 1993. "Review of NSFNET," NSF, Washington, D.C., March 23. 9. Some observers, for example, have raised concern about how the NSFNET coopera
FINANCIAL ISSUES 199 tive agreement was structured and implemented. These concerns were manifested in elec- tronic discussion groups over the Internet and eventually in congressional hearings in the summer of 1992 and an investigation by the NSF independent inspector general (which generally supported NSF decisions and actions) in 1993. 10. For this reason, there are a number of pilot projects being undertaken by cable and other providers apparently intended to gain a better understanding of consumer behavior and preferences. 11. One assessment, quoting several financial analysts, suggests that the combined winter rollback of cable rates by the Federal Communications Commission and the demise of the Bell Atlantic-Tele- Communications Inc. merger "will transform what had been a dynamic growth- and investment- oriented business into a more defensive cautious industry." It pointed to the prospect of difficulty generating the cash to invest in building the information superhighway and expressed skepticism about potential demand for consumer and entertainment-oriented services over it. Gilpin, Kenneth. 1994- "Market Place: A One-Two Combination Staggers the Cable Television Industry," New York Times, March 7. On the telephone company side, Regional Bell Holding Company bonds lost some appeal as investments because of bond buyer concerns arising from the current turmoil and confusion in telecommunications (although some analysts suggest that those concerns may be overblown). Hardy, Eric S. 1994. "The Superhighway's Slow Lane," Forbes, March 14, p. 136. 12. The Internet does not include settlements and charges associated with the provision of universal access subsidies, which are found in telephony. On the other hand, consumer advocacy groups often allege that the magnitude and burden of some telephony cost elements are both hard to sort out and sometimes reflect arbitrary accounting conventions. 13. That larger community will embrace educators, researchers, and their associates through links to their homes as well as their employers or schools. 14. MacKie-Mason, Jeffrey K. and Hal R. Varian. 1993. "Pricing the Internet," prepared for a conference, Public Access to the Internet, John F. Kennedy School of Government, Harvard University, May 26-27, 1993, November 26 version. Note that there are also some labor costs associated with the support of those facilities and the relatively low cost of a Network Operations Center. An illustration of how capabilities relate to costs may be found in experiences with NSFNET. The expansion from T1 to T3 service increased costs "dramatically," with $7 million of the $8 million in additional cost to upgrade service at the first eight sites attributed to the direct costs of transmission services (circuits, equipment, installation, and maintenance). "The three primary reasons for this cost increase were the new high-performance router technology, the costs of the additional capacity and speed on the MCI lines, and the requirement that the local lines [between internal and external switches] be able to handle T3 speeds." Office of the Inspector General, National Science Foundation, 1993, "Review of NSFNET." 15. Mile count of 16,200 from Jordan Becker, Advanced Network and Services, personal communication, April 3, 1994. 16. In the digital service hierarchy, the first step in this direction would be service at 1.5 Mbps. 17. Historically ISDN commanded a substantial premium, but that is changing in several areas. 18. There are also philosophical reasons in support of a private-sector approach, which was initiated under previous Republican administrations. 19. Snyder, Thomas D. 1989. 1989 Digest of Education Statistics. National Center for Education Statistics, Washington, D.C., December; U.S. Department of Education, National Center for Education Statistics (NCES). 1993. The Condition of Education, 1993. NCES 93-290. NCES, Washington, D.C., June.
FINANCIAL ISSUES 200 20. There is no general mechanism for guaranteeing that research (or education) applications that require the most advanced capabilities will be able to obtain them when or where needed. Regulations typically prevent the telephone companies from contributing (donating) the advanced new services while under tariff, and the window of opportunity for untariffed services is limited to very short periods for technology testing and experimentation. Those sites that do not have adequate connections in place will usually be unable to pay for them on an individual case basis as the charges will usually be much higher than when a planned buildout occurs for major portions of a community. Finally, sites that are in rural areas or those locations for which no good economic rationale can be made for a buildout may forever be disadvantaged without administrative remedies to offset the extra cost. 21. However, MIT also notes that its charges for connecting to the Internet via its regional network, NEAP, net, could be an order-of-magnitude higher if either the nearest point of presence were in New York or the access fee were assigned to NEARnet members on the basis of usage. Other campus computing and communications managers queried informally by the committee presented similar degrees of uncertainty about costs. Personal communication, James Bruce, MIT, February 1994. 22. Obviously, the practicality of commercial Internet service that reaches most people's homes depends critically on the nature of the last-mile connection (see Chapter 2), in particular, the way in which the cable TV and local telephone networks evolve and interconnect. Unfortunately, recent plans and announcements have revolved around a form of interaction in which the traffic from customer premises requires only limited bandwidth. Thus there is a fear among Internet users that the new facilities will fall short of what is needed for Internet-like services and, by extension, for an optimal, effective NII. 23. Most cable TV networks broadcast analog video. They are not designed to make individual connections to homes and have no operating reverse channel. So interactive data services such as the Internet cannot use the cable TV network today in general. 24. See, for example, Continental Cablevision. 1994. "Continental Cablevision, PSI Launch Internet Service: First Commercial Internet Service Delivered via Cable Available Beginning Today in Cambridge, Massachusetts," news release, March 8. This service will involve a metropolitan area network on Continental's fiber trunk lines, which Continental characterizes as an "off-MAN subscription to the Internet." More generally, a variety of special modems is being developed for use in trials for data communication over cable; programs have been announced by Prodigy and Cox Cable Communications as well as CompuServe Information Service and America Online Inc. See "Online Services Announce Migration to Cable TV; Move Driven by Low-Cost Zenith Homeworks Modem," pp. 8-9 in Information & Interactive Services Report, December 17, 1993. 25. Since the committee generated this idea, it has learned that such a study may be requested through telecommunications legislation under consideration in 1994. 26. Remarks by Stephen Wolff during a National Net '93 EDUCOM conference, Extending the Benefits, held April 14-16, 1993, in Washington, D.C. 27. Whereas it can be argued that businesses and other for-profit user organizations may also seek institutional connections, they are more likely to be able to provide the local infrastructure necessary to distribute access within an institution to the individuals that need it. 28. McClure, Charles R., et al. 1994. Connecting Rural Public Libraries to the Internet: Project GAINâGlobal Access Information Network, Project Evaluation Report prepared for NYSERNet, Inc., Information Management Consultant Services Inc., Manlius, New York, February 15. 29. For this reason, it is sometimes suggested that qualified outsiders be given access to institutions with adequate local infrastructure during periods of otherwise low use.
FINANCIAL ISSUES 201 30. There is a trend toward more affordable Serial Line IP (SLIP) access for some local public institutions such as high schools. 31. McClure et al., 1994, Connecting Rural Public Libraries to the Internet. 32. The original authorization limit for NSFNET approved by the National Science Board was $14 million; it was increased in June 1989 to $20 million based on the increase in traffic from September 1987 (75 million packets per month) to February 1989 (600 million packets per month) and the increase in number of institutions connected to regional networks. In May 1990 the National Science Board approved an increase in the authorization limit to $28 million in connection with the upgrade to T3 service, a limit that is expected to be met. See Office of Inspector General, National Science Foundation, 1993, "Review of NSFNET." 33. The transition will be eased by the planned gradual decline in support for regional networks. 34. For example, MIT's internal costs may exceed its payments to NEARnet by an order of magnitude. 35. Fisher, Lawrence M. 1994. "Reining in the Rising Hidden Costs of PC Ownership," New York Times, March 27, p. F10. 36. The package included a Macintosh Color Classic computer, a LaserWriter printer, an external CD-ROM drive and cable, and a computer-fax modem for hardware; a multipurpose software package; communications software; and a graphics/video-handling package (QuickTime Starter Kit). McClure et al., 1994, Connecting Rural Public Libraries to the Internet, p. 7. 37. The prominence of local facilities in the Internet-based research and education context points to a fundamental contrast between the telephony and NREN models of communication: the telephone company model employs intelligence in the network (owned by the provider) and assumes "dumb" user facilities; the NREN model assumes a relatively "dumb" network and intelligent user facilities. These models are reflected in the approach taken to users or customers: somewhat oversimplified, the telephone company customer is a consumer, and the NREN customer is a participant (who, for example, puts his or her material on the network for others). 38. Pricing for information services is structured in different ways, and pricing structures are volatile. Some are more subscription oriented (one up-front fee covers usage over a period); some are largely by the transaction, although there may be additional start-up and/or annual or monthly fees. Transaction costs may be by the page (e.g., $1.00 to $2.00 for LEXIS-NEXIS) or by the record (e.g., $0.12 to $30.00, depending on the database, for Dialog). There are also charges for connect time in some instances, which may range from less than $1 to several dollars per minute. 39. According to Roger Noll, "Charging higher prices to users with more intense demands [for information products], and lower prices to users who might otherwise be excluded, maximizes the diffusion of the information product." Noll, 1993, "The Economics of Information: A User's Guide," p. 36. 40. Congestion control mechanisms inside the network can push back on the traffic source dynamically as congestion is detected inside the network. The Internet, which lacks any pricing controls, depends on these push-back techniques to control offered load and limit congestion. The techniques are imperfect, however. Not only is the congestion control algorithm not enforced by the network, but it is also not even part of the protocol architecture of applications that do not use TCP. UDP and video traffic have no current implementation of this, and in fact are "unfair" to the hosts that do implement congestion control. 41. PL 102-194, Section 102. 42. However, the analogy with a freeway system is imperfect, because it takes little
FINANCIAL ISSUES 202 effort to set a computer on an infinite network capacity-consuming task, whereas highway use demands time from the driver. 43. MacKie-Mason and Varian, 1993, "Pricing the Internet." 44. MacKie-Mason and Varian, 1993, "Pricing the Internet," p. 16. 45. To some observers it seems that paying users will not be happy with best-effort service. Experience suggests that user satisfaction depends on the quality of service delivered, in which case it is argued that usage-based charging is the efficient approach. Unknown is how strong that relationship may be, and what it implies for consumer response to alternative pricing strategies. 46. See Shenker, Scott. 1993. "Service Models and Pricing Policies for an Integrated Services Interact," June 8 version, prepared for a conference, Public Access to the Internet, John F. Kennedy School of Government, Harvard University, May 26-27, 1993; and Cocchi, Ron, et al. 1993. "Pricing in Computer Networks: Motivation, Formulation, and Example," August 30. 47. In retrospect, pricing mistakes may have contributed to poor market performance for X.25 and ISDN offerings. 48. Interestingly, however, in the consumer market, Prodigy and GEnie were reported to have abandoned flat-fee for metered systems based on connect time in early 1994. Vola- tility in pricing appears to be the norm in network-based services, given imperfect understanding of nonconstant consumer behavior and its impacts on network load and on revenues. See, for example, Lewis, Peter H. 1994. "A Traffic Jam on the Data Highway," New York Times, February 2, pp. D1, D5. 49. The committee recognizes that price and cost often seem to have had a tenuous relationship in communications. For example, recent public discussion of the apparent costs of sending a fax by Internet versus over the telephone system points to the fact that the phone system allocates more costs more fully than does the Interact. Although the phone system entails some extra costs associated with mandatory cross-subsidies, the true magnitude of those costs is subject to debate, as is any particular allocation of costs. 50. CSTB, 1988, Toward a National Research Network. 51. Shenker, Scott, David D. Clark, and Lixia Zhang. n.d. "A Scheduling Service Model and a Scheduling Architecture for an Integrated Services Packet Network"; Cocchi, Ron, et al., 1993, "Pricing in Computer Networks: Motivation, Formulation, and Example"; and Estrin, Deborah, and Lixia Zhang. 1991. "Design Considerations for Usage Accounting and Feedback in Internetworks," IFIP International Conference on Integrated Network Management, April, pp. 719-733. 52. For example, those with fewer resources might have to rely more on longer-term prepaid arrangements, while those with more resources would have more flexibility to afford spontaneous and more expensive arrangements, and so on. 53. Specifically, there may be an analogy to the kinds of access charges seen in the telephone system that combine basic network access in the local loop with usage charges for long-haul or special services (e.g., the Exchange Network Facilities for Interstate Access tariff and its progeny). 54. Note that the High-Performance Computing Act of 1991 provided that. "[a]ll Federal agencies and departments are authorized to allow recipients of Federal research grants to use grant moneys to pay for computer networking expenses." PL 1021-94, section 102(f). 55. AT&T and other companies have been reported as favoring some form of targeted individual assistance delivered along the lines of "telephone stamps" intended to assure that benefits are provided to genuinely low-income people. Pearl, Daniel. 1994. "Debate Over Universal Access Rights Will Shape Rules Governing the Future of Communications," Wall Street Journal, January 14, p. A12. 56. MacKie-Mason and Varian, 1993, "Pricing the Internet," pp. 16-17.
FINANCIAL ISSUES 203 57. The network is a general-purpose vehicle for many services. Information services are more specific, begging the question of which services might be eligible for coverage in a financial support program. 58. Einhorn, Michael A. 1993. "Toward Greater Achievements: New Policies for the Internet." 59. Noll, 1993, "The Economics of Information: A User's Guide."