L. Jean CAMP <firstname.lastname@example.org>
Rose P. TSANG <email@example.com>
Sandia National Laboratories
Why do Americans lose their phone service? What functions of protocols could support or undermine universal service? Why don't all Americans who are passed by cable television subscribe? If communications protocols were optimized for universal access, what characteristics would such protocols have? How do proposed Internet services compare with a theoretically optimal protocol with respect to universal service?
Pricing for packet switching has focused on economic efficiency (e.g., Mackie-Mason, 1995; Choi, Stahl, & Winston, 1997; Shapiro & Varian, 1998), billing to encourage widespread adoption of network innovations (e.g., Xie & Sirbu, 1985) and billing in a manner consistent with the underlying network (e.g., Clark, 1996). Here we look at billing Internet services with respect to universal service.
Universal service has traditionally been based upon two fundamental assumptions: that there exists a single service to which all are entitled and providing this service is a function of geography. The single service to which all are entitled, plain old telephone service, is local voice access bundled with traditional toll service (i.e., long-distance service). The first assumption is increasingly meaningless, as telephony and data networking merge. The second assumption is flawed in cases where the infrastructure investment has been recovered, or is wireless.
The bundling assumption is an increasingly irrational selection of exactly two data services out of the many. Consider, for example, that text-based access to governmental data and job openings may be more valuable than voice toll service. Yet, despite its flaws, any universal access regime will be built upon the voice-based universal service regime. While the technological assumptions of universal service are dated, citizen access to information and communications is becoming increasingly important. With the rise of xDSL and the continued popularity of telephone-based modems, telephony's universal service may be the backbone of the information age's universal access.
We begin by describing telephony's mechanisms for funding universal service through inter-customer, inter-company, and intra-company payments. Historically, a constant quality of service (QoS) has been in conflict with universal service. We argue that the future problems with universal service will not be the simple penetration rates problems of today, but will be more subtle issues of availability and QoS. (The penetration rate is the percentage of households with POTS; it is broken down by income, race, age, geography, and family structure.) We show that the current regulatory mechanisms are inadequate for a competitive market where a continuous variable, QoS, not a Boolean variable, existence of service, is key.
We then discuss protocols that include mechanisms for QoS, and consider how the services available with each would influence universal service. We focus on how these new QoS protocols for packet-switched networks, and the pricing structures these protocols enable, can provide universal service in digital networks. We show how implementation of these protocols in a competitive environment can result in broader access to communications service than is the case today.
Universal service has failed poor families in urban areas; the young more than the old; and native speakers less often than those who speak English as a second language. The profile of those without telephone service is described in greater detail in Section 2. Their reasons for not having service are primarily that toll pricing is perceived as random, and an initial deposit is required. We discuss secondary issues for loss of service and how advanced services could help, or aggravate, the underlying problems. We illustrate how different modes of operation could support pricing models that address the continuing failures in the universal service regime. We argue that the critical components needed to support universal service are end-user feedback, entry controls, and support of best-effort traffic as well as guaranteed service traffic.
Telephone service providers use explicit, delayed feedback to influence user behavior. Through pricing, providers attempt to offer only as much as their networks will bear. By requiring deposits, they limit their exposure to customer abuse. Through pricing differentials, companies attempt to level network use and succeed in increasing pricing uncertainty. For example, the need for a deposit illustrates how telephone companies' needs to limit risk and the consumers' need for service have been in conflict. QoS and bandwidth reservation technologies combined with best-effort service can resolve some of these conflicts.
Telephone companies can provide real-time feedback to users about service availability; for example, through denial of service as soon as a specific call reaches a given cost threshold, or at some user-selected monthly total charge. The potential for denial of service at the entry point would remove the need for a deposit.
Companies can provide highly affordable, best-effort service at a consistent price. This would level network usage in a real-time manner and remove consumer uncertainty. Today companies use rough measures for providing price incentives in time periods other than peak, which creates uncertainty for less informed consumers. Companies build for peak to ensure that highly profitable corporate data do not encounter congestion. Best-effort service could provide affordable, constant rate, toll access off-peak. There would be short-term denial of service as opposed to long-term loss of service. Publicly available AT&T data suggest that this denial of service would occur only between 9 and 10 a.m. weekday mornings. Currently companies can, and some do, offer a maximum monthly charge yet the customer can neither negotiate this maximum with any degree of granularity nor negotiate it periodically. The ability to control one's own bill would remove uncertainty and give families the ability to control their monthly expenditure on communications services. All of these services and changes are possible with any set of protocols that provide end-user feedback, entry controls, and support of best-effort traffic.
We argue that usage-based pricing can be more compatible with universal service than flat rate is, in contrast to Anania and Solomon (1988). Such a usage-based system must provide mechanisms that remove uncertainty for the consumer, unbundle services, and provide support.
These services point to promise in providing universal service. Convergence, as a technical reality grounded in well-defined protocols, will alter universal service. We take an optimistic approach, but our optimism is supported by the history of telephony penetration, the practice of building for peak loads, and the capacities of next-generation networking protocols.
The Internet and telephony were traditionally based upon very disparate engineering technologies (packet switching and circuit switching, respectively) as well as very different types of service offerings. The Internet has used best-effort data traffic for e-mail and file transfers, and telephony has used guaranteed low bit-rate circuits for voice traffic. The convergence of these networks has resulted in the proposal of packet-switched networks containing protocols that provide quality of service (QoS) mechanisms. The economics of the resulting packet network may violate the fundamental assumptions on which universal service funding is built. Thus a new consideration of universal service is needed for next-generation networks.
Universal service is targeted at two groups: rural subscribers who may be unable to support the infrastructure necessary to provide service and the urban poor. Wireless technologies have made the issue of wired infrastructure, and thus considerations of distance, increasingly unnecessary. As the politics of universal service is not at issue here, we will focus on the poor and not consider distance issues.
What is needed is a universal service model for a packet-switched network that supports both best-effort traffic and traffic with guaranteed quality of service. It would not be ideal to have a new paradigm for universal service based on packet-switched networks with only best-effort service developed just in time to be obsolete in the face of next-generation networking.
The concept of universal service came out of the agreement between the United States government and AT&T, then headed by Theodore Vail. All other nations had or were in the process of nationalizing their telephone systems as an extension of the postal system. In the United States, however, despite a short interlude of government ownership (Brooks, 1975), AT&T became a regulated monopoly.
In fact, AT&T has never been the only phone company in the United States. Even after the creation of the Federal Communications Commission, small rural phone companies existed (with which AT&T was required to interconnect following the Kingsbury Commitment). There is some argument over whether universal service was more effective pursued under a regulated monopoly with competition. There is no debate that universal service with a single quality of service did not exist until after implemented by AT&T. Rural cooperatives had low qualities of service both in the times of provision and signal quality. In a farmer-owner cooperative the owners ran the switchboard in between other responsibilities, line repairs were made when possible, and phone lines were often wire fencing doubling as a transmission medium (Fischer, 1992). Thus, quality of service has historically been inversely related to the widespread adoption of services.
The universal service fund is actually a misnomer; rather than being a fund, universal service is supported by series of exchanges and payments made between carriers. Long-distance carriers pay local exchange carriers, with the assumption that long-distance calls are dominated by the affluent subscribers who can afford the subsidy. Universal service has been implemented through separations and settlements. Separations are the payments made by one company to another; settlements are the payments made between divisions of the same company.
The payments subsidize basic services, qualifying individual households, and qualifying institutions. The universal service fund has been expanded to include connecting schools and libraries to the Internet. In practice, every qualifying school or library can obtain connections for Internet services for less than the market rate. This is the first extension of universal service beyond common carriage. Previously only the mail and telephones came under universal service. Radio and television have not been subject to universal service in the United States; however, this is not true for all nations. For example, in the Soviet Union and China, instead of people having individual radios, loudspeaker systems piped news to all common areas. In West Germany everyone has the right to a radio for the public safety purpose of emergency planning and response. Yet in the United States universal service effectively continues to mean plain old telephone service with no subsidy for cable penetration, wireless penetration, or even availability of pay phones.
In the previous section we described the goals and philosophy of universal service. In summary: Telephone service is subsidized for a targeted population. Broadcast services are not subsidized. Internet services are subsidized for qualified educational institutions. In addition, postal services are subsidized for rural households and institutions by constitutional requirement.
The most complete study of universal service losses to date (Mueller & Schement, 1996) shows that loss of phone service can be predicted by income, ownership of assets, and age of the head of the household. In fact, age of head of household predicts that young families are the most likely not to have phone service.
The primary reason for loss of service is a high toll-call bill. Toll-call costs are unpredictable unless the caller understands the charging mechanism and can control all the calls made by all members of the household. Local call charges are billed periodically, not usage-based, and provide monthly feedback about costs. Toll calls are increasing flat rate, usage-based and continue the historical pattern of providing monthly feedback about pricing. Historically, toll-call rates have been based on predicted congestion and have been priced by time of day. Toll calls were also historically priced according to cost of the infrastructure, and thus vary by distance. While time charges are not uncommon (e.g., 5-cent Sundays) telephony pricing mechanisms are becoming increasingly based only on duration of the connection.
Consider billing as economic feedback. Phone calls are billed monthly thus providing periodic pricing feedback. The decision to accept a connection or make a specific call is not directly linked to the payment for that call.
Other common reasons for loss of phone service are large calling card or collect call charges. In all three cases the charges can be made by parties who are not responsible for the phone bill. In the case of collect calls and calling cards the cards may be on a different provider than the one selected by the party responsible for phone service. This results in a remarkable lack of price predictability. All these observations suggest that mechanisms that reduce uncertainty will reduce the incidence of loss of phone services. Mechanisms that reduce uncertainty can either create a constant bill for service, thereby reducing variability, or provide real-time feedback about costs.
An unpredictable feature of loss of phone use is that the poor use more expensive telecommunications services per dollar of income than users in any other quintile. Hispanic and African American households spend more on cable TV, long distance, and advanced services (e.g., call waiting) than white households do. Certainly the resulting higher rates of disconnection, when linked with lower computer penetration, undermine claims to Internet democracy and point instead to higher divisions of haves and have-nots.
Consumption of expensive telecommunications options is one reason for disconnection. This suggests that to limit overall phone bills would serve universal service. Note that the same groups underserved by credit markets are conversely overserved by telecommunications markets, which function on the credit model.
Other features that might characterize a protocol compatible with universal service include unbundling entertainment from calling, concentrating control in the hands of the party responsible for the phone bill, and creating protocols with a low information threshold. These would support universal service regardless of the technical mechanism used.
Consider that homes without telephones often have cable television, even premium channels. Thus subscribers can manage their cable bills even though overwhelmed by the phone bill, as evident by disconnection of phone service. Cable television provides valuable services, including new programming. In neighborhoods where children are effectively under house arrest because of high levels of street violence and open drug trades, entertainment may provide a more important value than connectivity. Some proposals would link not only tools and local service, but also broadband entertainment access.
Concentration of control in the hands of the party responsible for the bill will increase communications penetration. Calling cards, long-distance calls, and collect calls enable anyone with access to the physical device to charge calls, thus enabling adolescents, irresponsible guests, and relatives to burden the phone owner with uncontrollable charges.
A low information threshold has not been previously identified as a driver of universal service. However, interviews with those who have lost phone service illustrate that uncertainty in billing is a problem.
A previous work on universal service characteristics focused upon the structure of the regulatory regime (Gillett, 1994); it argued for a regulatory approach to universal service that is compatible with this set of technological specifications. This approach would make explicit subsidies available directly to low-income users, not define immature services as essential, ensure competition, and finally favor technologies that are digital, scalable, and extensible. In contrast we argue that digital, extensible, and scalable technologies are not all created equal. In fact we propose that there exist specific characteristics for QoS mechanisms which can undermine universal service, or conversely can increase the connectivity of the poor. Furthermore, the Mueller and Schement study showed that the services targeted specifically for the poor are not reaching poor young families. Thus the systematic causes of denial of service in a QoS network need to be addressed at a fundamental level in a network that provides both best-effort and QoS transmission.
We now consider three protocols proposed for charging for Internet traffic: expected capacity service (Clark, 1996), a smart market (Mackie-Mason & Varian, 1995), and RSVP (resource reservation protocol; Zhang et. al., 1993). We compare the features of these systems with those features that would support universal service.
In the previous section, we identified characteristics of protocols that would facilitate universal service. Such characteristics include network support for best-effort services as well as guaranteed services, reducing pricing uncertainty by providing feedback about costs, entry controls, more control for the responsible billing party, unbundling of services, and low user-information thresholds. In this section we discuss three QoS network protocols in light of their potential applicability to universal service. The first two schemes perform dynamic bandwidth allocation at the packet level. The third scheme performs dynamic bandwidth allocation at the flow level, that is, connection-oriented resource reservation.
The expected capacity service is a two-level priority scheme proposed by Clark (1996). It provides a guaranteed minimum capacity service with a guaranteed burst size allowance provided by a token bucket. Clark considers the scheme in terms of a simple sender-pays model or a more complex receiver-pays model. The sender-pays model is as follows. Each user selects a profile. The profile consists of two measures: minimum rate and token bucket depth (comparable to transmission control protocol's (TCP's) window size.) In response to the user's selected profile, the service provider offers a price for the time period (e.g., a month) in which the profile will be enforced. Thus users are provided a fixed and predictable price for their selected profile.
Users make connections and use network services without pricing each individual connection. At the edge of the network, the service provider tags each packet as either being "in" or "out" of the user's profile. At a congested network switch, packets tagged as "out" will be dropped. If no congestion occurs, packets tagged as "out" may be delivered thus providing the user with more effective bandwidth than their selected profile specifies. At nonpeak hours, this effectively allows users to consume resources, in a best-effort manner, irrespective of their selected profile without penalizing the performance of other users. This provides a potential mechanism for the distribution of network usage amenable to universal service.
Notice that having the receiver pay in all conditions means the sender must know of the receiver's usage profile to properly label packets. For example, if the initial sender requests a Web page, the responding server would be required to know the usage profile of the requester to know how to tag the requested item for the receiver-pays model to serve the client's request. To overcome this, Clark suggests that the request be sent with the usage profile so that packets may be tagged appropriately. To enable all complexities of profile to be sent would require active or executable usage profiles, which complicates the scheme remarkably.
This model for pricing the Internet provides the possibility of real-time user feedback. As packets are dropped or slowed, the users can be notified. If the users are not notified, they may perceive the slowed service and may understand this as indirect notification. The source of the slowed service may or may not be identified. The addition of notification of the users that their connection has changed priority and the reason for the change (total bandwidth out of profile, use at this time not in profile, etc.) would make this mechanism better suited for universal service.
Some complexity and thus the possibility of information overload is associated with this service. Pricing feedback should be directly available as the user is selecting the profile.
Authentication could allow only the party responsible for the phone to change the profile so that other parties cannot increase the bill. As the profile is altered periodically and then set for all device users for a given period, this protocol is amenable to user, as opposed to device, authentication.
In Mackie-Mason and Varian (1995) a Smart Market for responsive pricing in the Internet is proposed. In this model the user attaches to each packet a bid or amount that the user is willing to pay to get a specific packet through the network. When packets are dropped, those packets with the lowest bids are dropped first. Each packet with a higher bid is charged the amount of the lowest bid submitted into the network; that is, for each packet, the user is charged a threshold that is less than or equal to the amount bid. In the Smart Market approach, pricing is explicitly identified as a form of economic, as opposed to network, feedback. In this, we certainly agree with the authors.
The authors include in the discussion of the Smart Market the proposal that some users can choose to pay nothing and receive only best-effort service. The authors further note that they expect the Smart Market to be implemented with software so that users do not have to bid on every packet; the authors suggest that the users halt sending when the price goes above, for example, U.S.$0.0001 per packet.
By implementing the packet auctioning as an active response (not controlled by the user at the packet level) that observes user selections and adapts within user controls, this protocol could serve as effectively as the previously described expected capacity service. In this case an agent would search for optimal routes and bids, and the total monthly bill would be some predictable amount. This would also remove the high user-information overhead. As economic theory suggests, the importance of people's satisfaction may outweigh the benefits of optimization, so the decrease in user overhead may in fact not decrease the quality of user's expressions of preferences.
Conversely, when packet auctioning is implemented with an active user, those people bidding must understand the probability of congestion occurrences and the bandwidth demands their applications make on the network. Users do not know how much money was spent to complete a connection until after the connection is complete. Two transactions that seem identical to a user could have very different costs. Issues of receiver-pays mechanisms were not addressed. Again, when a sender pays only for the request while the party contacted pays to transmit an entire Web page, this mechanism may prove to be flawed.
This protocol is subject to information overload. Note that the need for the head of household to be able to control charges would require authentication for every packet initiated. If the total bill could be periodically set, as with a profile, then occasional user authentication would be adequate. This could be implemented with session authentication so that each user can spend a certain amount. Session authentication and user-specific pricing would increase information overload.
RSVP (Zhang et al., 1993 is a protocol designed for IP (Internet protocol)-based networks for setting up end-to-end QoS across a heterogeneous network. In RSVP resource reservation is based upon flows, that is, streams of data from a source to a particular destination. Each flow has an associated specification (flowspec) that contains information about the QoS of the flow. This information tells each router in the connections which resources the router should reserve for that particular flow. Thus when a user requests a flow setup, the flowspec is propagated to the appropriate branch in the multicast tree where, if each intermediate node is capable of satisfying the flow's requested QoS, the request is accepted. Once a flow is set up it corresponds to a logical dedicated circuit; the flow's QoS is guaranteed, regardless of potential congestion, until the circuit is explicitly torn down.
The initial motivation for RSVP was to support more efficient multicast operations. Thus it is based upon a receiver-initiated request model. In more traditional (non-receiver-initiated) multicast protocols, a sender would have to know the QoS requirements of all the receivers in the group (for example, receiver A requests 128 kbps, receiver B requests 2 Mbps, etc.). Clearly, this would require an exchange of a large amount of information between the sender and all receivers, as well as the maintenance of a large amount of dynamic state information.
As mentioned before, most billing situations correspond logically to the sender-pays model. Thus RSVP, being receiver-initiated, is naturally amenable to the receiver-pays model. By enabling receiver-initiated resource reservation, each receiver can reserve the resources it requires and at the same time be assigned a corresponding price for the specific QoS request. Each receiver needs to know only about its own QoS requirements. To support the sender-pays model, an extension to the RSVP protocol must be used where the QoS requests are propagated further up the multicast tree to the source node.
Two pricing methods are best suited for RSVP. Pricing may be provided as a fixed and predictable function of the user's requested QoS. The problem with this approach is that network resources become inherently more valuable as utilization, that is, potential congestion, increases. Thus the price charged would not necessary reflect the immediate "value" of the resources. The other way to price a RSVP flow could be as a function of both the requested QoS flow and the network load. However, the problem with this approach is unpredictable pricing for users. Users may not know when they were using more "expensive" network resources, such as during peak hours, until after the resources were consumed.
Since in RSVP each application's flow requires a separate flowspec, the unbundling of services follows naturally. However, since users must be able to specify the QoS requirements for each application's flowspec, information overload may be associated with RSVP. A possible solution is to provide direct pricing feedback as the user is selecting the parameters for the flowspec.
Authentication should allow only the person who is responsible for network charges to change the flowspec. As with the Smart Market this can be implemented with session authentication or by periodically setting a total bill and letting the software determine the requirements.
In conclusion, we explain the impact of the features of the previously discussed QoS protocol mechanisms on the viability of universal service.
Table 1 lists the characteristics of the QoS mechanisms examined in this work. While the protocols vary with respect to the ability to predict, and therefore control, price, all protocols are built with the assumption of convergence. If convergence results in the bundling of currently separate applications, such as pay television and telephony, universal service is expected to decrease.
|Expected Capacity||Dynamic Auction||RSVP|
|Unbundling||Through profile||No||Through flowspec|
From an examination of the above protocols, a trade-off appears in predictability in quality and predictability in price. Predictability in quality requires reserving network resources regardless of the presence of congestion, and thus regardless of the unpredictable actions of others. Predictability in price requires cost-controlling response to scarcity, which leads to variable QoS delivery.
Table 1 examined characteristics that, as we argued in Section 2, affect universal service at least in terms of telephony penetration. In table 2 we consider the technical mechanisms that support the economic characteristics favorable to universal service.
|Expected Capacity||Dynamic Auction||RSVP|
|Real-Time User Feedback||Possible||Possible||Possible|
|Differentiation of Applications||Possible||No||Yes|
|Authentication Needed for Payer Control||To alter profile||Session/User||Session/User|
The Internet, or IP-based services, have promise because of technical characteristics that provide the potential to expand universal service: specifically, the ability to offer flat rate best-effort service, the potential for authentication before adding a usage-based charge, the potential for unbundling at a detailed level, the ability to offer pricing without requiring an understanding of the underlying billing mechanisms, and finally, the potential for customized service level with specific monthly charges selected in advance. All these features work toward the goal of true universal service.
End-user feedback may be useful in providing universal service if real-time cost information enables users to control their bills, thus removing billing uncertainty. In addition, the use of entry controls can assist in universal service; entry controls enable networks to either refuse entry after reaching some trigger point, or refuse services for which there would be further charge. These features would remove all uncertainty and help to control bills. The ideal use of best-effort traffic with respect to universal service is to allow for low-cost service without disconnection when subscribers have overspent for premium services.
Before the deployment of any significant network mechanism such as the previously discussed QoS mechanisms, it is crucial to understand the mechanism's impact on the viability of the services it may be expected to support. As seen from experiences with telephony networks, as well as the Internet, once a specific networking technology is deployed, its impact on subsequent service offerings is indeed lasting.
Current forecasts of business mergers and regulatory actions suggest that more of the hazards of Internet accessibility will be met than the promise. Thus it is crucial that universal access be considered in the design of protocols for the next-generation Internet. This paper provides the information for protocol designers and early adapters to develop and select protocols that can serve universal access.
Anania & Solomon, 1988, User arbitrage and ISDN, Intermedia, London, January.
Brooks, 1975, Telephone: The First Hundred Years, Harper & Row, New York.
Clark, 1996, Explicit Allocation for Best Effort Packet Delivery Service, The Telecommunications Policy Research Conference, Solomons Island, MD.
Choi, Stahl, & Winston, 1997; The Economics of Electronic Commerce, Macmillan Technical Publishing, Indianapolis, IN.
Fischer, C., 1992, America Calling: A Social History of the Telephone to 1940, University of CA Press, Berkeley, CA.
Gillett, S.E., 1994, Technological Change, Market Structure, and Universal Service, Telecommunications Policy Research Conference, Solomons Island, MD, October.
McKnight, L.W., & Bailey, J.P., 1997, Internet Economics, MIT Press, Cambridge, MA.
Mackie-Mason & Varian, H., 1995, "Pricing the Internet," Public Access to the Internet, ed. Kahin & Keller, Englewood Cliffs, NJ: Prentice Hall.
Mueller & Schement, 1996, Universal Service from the Bottom Up: A Study of Telephone Penetration in Camden, New Jersey," The Information Society, Vol. 12, pp. 273-292.
Shapiro &Varian, 1998, Information Rules, Harvard Business School Press, Cambridge, MA.
Xie. J & Sirbu M., 1985, "Price Competition and Compatibility in the Presence of Positive Demand Externalities," Management Science, June 1995.
Zhang, Deering, Estrin, Shenker, & Zappala, 1993, "RSVP: A New Resource ReSerVation Protocol," IEEE Network Magazine, September 1993.