Objections to responsive pricing

We explore some common arguments against responsive pricing in network operations in this Section, and provide some counterpoints. Some previous work along these lines is contained in [e] and [10].

• Why do we need to introduce prices? The Internet is free now — let's keep it that way.

  • Counterpoint: The Internet is not free now, though it seems that way to many users whose universities or organizations pay the access fees. The issue is not whether Internet usage should be priced: it already is. The issue is how the Internet should be priced so that its value to the users is maximized.

• Network resources will soon be essentially free, therefore Internet congestion and the accompanying QOS degradation will not be a problem in the future.

  • Counterpoint: This represents an optimistic view of the future, but we do not believe that this will come true in the short or medium terms, if ever. See our arguments on why efficiency will continue to be important in Section 2.

• If congestion is caused by bandwidth-intensive users such as real-time video, why don't we just keep these users off the Internet, or limit their number so that they don't cause congestion problems?

  • Counterpoint: Keeping these users off the Internet means keeping the Internet low-tech and continuing the best-effort no-guarantees paradigm. This runs contrary to the trend towards integrated-services networks, and may cause the Internet to miss out on innovative information transfer and retrieval mechanisms. Apart from the administrative issues, why should ''low-tech'' users be allowed to veto "high-tech" users? What will members of the general public want when they come online? Which administrative bodies do we want to empower with rationing authority?[e]

• Why won't some non-pricing scheme be enough? Administrative controls can be used to impose some appropriate notion of fairness, for example; or users can choose a traffic priority level which matches their requirements.

  • Counterpoint: Who decides what is fair? The network operator can; but according to a user-oriented objective, fairness should be determined collectively by the users. We might all agree that telesurgery is more important than email, but what about interactive video games versus email? Also, every time a new application is developed it has to be slotted into the priority order, an increasingly complex process. Further, the value of an application to a given user will vary over time. Priorities for different applications will sometimes — perhaps usually — incorrectly order valuations. Suppose the network simply supports priority levels and allows each user to choose their own level. Why wouldn't they all choose the highest priority? To guard against such abuses, there would have to be some penalty for "inappropriate" declarations, implying the need to define "appropriate" priority levels or to assign increasing charges to higher priorities (see e.g. [2]). A user's choice of priority level would then be based on economic considerations: balancing the benefits of higher priority against the costs and/or the penalties for inflating their application's perceived priority level. This is the essence of a pricing scheme.

• Most users will want to know their charges in advance, and will not want to deal with prices that change during the lifetime of a typical connection. Why won't flat-fee prices (per minute connected, or per kbps of the access link) be enough?

  • Counterpoint: We are not advocating that all users must face responsive prices. Any user can choose not to face dynamic prices, even if their application is adaptive. They would then be charged according to some other pricing scheme. For example, a user might be allowed to pay zero responsive prices in exchange for getting only best efforts service with lower priority than other users who pay a positive price. In any realistic pricing scheme it would be possible for a user to set the maximum charge they are willing to pay, which is what is usually required for budgetary purposes.

    We should also point out that flat-fee pricing is really long-term usage pricing, so even "flat" fees include a usage-based component: it's just averaged over a period much longer than a connection lifetime.[f]

• Bits/bytes/cells/packets are not the correct units to charge for — it's information that users care about. Any scheme that proposes to look inside every packet to determine how it relates to other packets is likely to be too complex to be justified. Also, lower-layer mechanisms (such as Ethernet collisions) or packet losses requiring retransmissions make it difficult to predict how much "raw" data has to be transferred to transmit a given amount of information. Should users be charged for retransmissions that they have no control over, or packets that are dropped by the network?

  • Counterpoint: Our proposal involves pricing for transport, not for content. The ''importance'' of a particular packet, and its relation to other packets, is a higher-layer issue determined by the application (or ultimately by the users). We are not proposing that the network be aware of these issues; on the contrary, with responsive pricing it's up to the users to decide how packets are used to transfer information. It's true that it is in general impossible to determine beforehand exactly how many packets are required to transmit a block of information, but again this is a higher-layer issue. Indeed, there will be some efficiency gains from providing a financial incentive to software developers to make more efficient use of network packet transport. The important question is whether the users or the network should bear the uncertainty arising from variations in congestion. If the network is expected to offer a ''file transfer'' service, the file transfer charge per megabyte could be computed by averaging over many such transfers. If the user is expected to pay for all transmitted packets, they could define a maximum number of packets they are willing to transmit per megabyte of information, and invoke an application-layer process if this threshold is exceeded.

• Dynamic pricing schemes are unworkable in practice due to the overheads involved in accounting and billing for usage on such a detailed level. In addition, a significant portion of the revenue raised is needed to defray the cost of doing dynamic pricing in the first place!

  • Counterpoint: The costs of dynamic pricing may outweigh the benefits for a particular implementation but we do not believe this is necessarily true for all dynamic pricing schemes. In particular, online pricing mechanisms may reduce the actual cost to an acceptable level; there is no reason to think that current billing and accounting costs in other industries, such as telephone or electricity networks, will necessarily apply to dynamic pricing in the Internet. In particular, since data networks have vast distributed computing power in the form of smart end-user devices at the periphery, it may be possible to design distributed billing systems that have very low cost for large numbers of small transactions.

• Dynamic pricing is impractical because users cannot respond to prices that are updated many times per second. If the update interval is increased to the minimum period in which users can respond, congestion can arise and disperse in between price updates, so that prices no longer influence user behaviour.

  • Counterpoint: Our scheme assumes an intelligent network interface at price-sensitive user sites, so the processing necessary to respond to dynamic prices would be done automatically based on pre-programmed user preferences. For example, a user could have a default preference in her email program that instructed the software to hold outgoing email whenever the price exceeds 0.01 cents per packet. Such software would play a similar role to current TCP implementations, which respond to network feedback by adjusting their traffic inputs, except that the feedback in our case is the current price.

• Once a network is installed, any load-dependent costs of transferring data are minimal — the fixed costs of network management and maintenance dominate. These fixed costs can be efficiently recovered through connection fees and capacity prices (proportional to the size of the access link). Why implement an elaborate pricing mechanism to recover the relatively small variable costs?

  • Counterpoint: The point is not current cost recovery. We are concerned about the congestion cost that one user's traffic imposes on other users sharing the resources. Bandwidth or buffer space occupied by one user's traffic is not available to other users. When this reduces other users' quality of service (through increased delays, loss rates, blocking probabilities, and so on), they suffer congestion costs which may translate into significant actual costs of service degradation. One mechanism to capture these costs is a price that is sensitive to some indicator of congestion, such as load.

• Even if we want to allocate according to congestion costs, how can the network determine what actual costs the current load is imposing on users who probably have widely varying service requirements? Getting users to reveal these costs is likely to be extremely complicated, if not impossible.

  • Counterpoint: It is true that providing users with the right incentives to reveal their actual costs of service degradation is complicated. It is not impossible however: truthful revelation is one of the properties of the smart market mechanism in [6]. With any prices that increase with the degree of congestion in the network, users will be induced to prioritize their traffic. Only users who value their traffic at least as much as the current price will transmit. If congestion remains unacceptably high, then the associated price was too low; conversely if capacity is unacceptably underutilized, the price was too high. Thus, through a process of experimentation and dynamic adjustment, the network can shape the price schedule so that users approximately reveal their valuations for uncongested service through their responses to the price feedback.

• Charging for transmission fails to capture cases where the benefit of a transfer is with the receiver. If senders are charged for receiver-initiated transfers, we could see a drastic reduction in the number of open-access servers with a corresponding decrease in the value of using the network.

  • Counterpoint: The problem of allocating the benefits of a particular information transfer is a higher-layer issue. We do not believe that associating the charge for a transmission with the sender constrains the actual flow of money in any way. It is easy to imagine multiparty connection protocols which initially negotiate each party's responsibility for the total charge, or "reverse-charges" servers which only transmit data once the receiver has indicated willingness to pay the resulting transmission costs. Just as in telephony, we can expect "1-800", "1-900" and other billing services to arise.

• Suppose we institute some form of responsive pricing, and users (especially the high-bandwidth ones who will pay the most) leave the Internet and use other networks. Won't that reduce the value of being on the Internet, perhaps to the point where even small users leave and join the other networks?

  • Counterpoint: We are discussing only charging users for the amount of congestion cost they impose on other users. Users get to decide whether they want to pay money to avoid congestion, or not pay money and bear congestion delays — in the current Internet, everyone is forced to accept the latter alternative. Costs are not just monetary: if the cost of congestion delay is severe, then we can expect that some users are already being driven away. Indeed, many network applications are restricted to private leased-line networks to avoid Internet congestion (e.g. most videoconferencing). By allowing transport priorities to be sorted based on who suffers the most from congestion delay, we will increase the value of the network, which should spawn additional growth and new uses.

• Why are we so concerned with modifying individual user behavior anyway? Surely one user can't do that much damage to the Internet?

  • Counterpoint: One user, or a relatively tiny number of users, can now do a lot of damage to the Internet. A single interactive video connection can take up as much bandwidth as thousands of traditional Internet applications. Without some incentives to take other users into account (and/or penalties when they do not), a small fraction of the user base could bring large regions of the Internet to a standstill. In any case, the collective behavior of lots of individuals, acting without concern about the effects of their traffic on others, can easily lead to congestion. We think it most natural and efficient to attack the problem at source, but it may be that feasible responsive pricing schemes are more practical if imposed at a higher level of aggregation.

• There is already a penalty for heavy network usage: my application runs slower. Why should I pay again, in real money?

  • Counterpoint: Your application running slower represents a penalty to you, but what about other users' applications which are also running slower? In order to efficiently share resources, you have to be made aware of the costs your usage imposes on other users. If there is enough of the shared resources, these congestion costs can be insignificant. But we believe that the Internet cannot rely on these costs being essentially zero, at least not for the foreseeable future. Meanwhile, some forms of responsive pricing give the user a choice: either pay in delay, or pay in money to avoid delay. This choice is available on a gross scale today: we can use the Internet with uncontrolled delays, or use a low-delay private leased network. We think it is possible to offer this pay-or-delay choice to users within the Internet, making them better off by giving them a wider range of service choices.

• With any form of responsive pricing, it's the small users who will suffer the most. Rich users could behave as they want since they have the resources, and could effectively limit the network access of smaller users. The role of the Internet as a medium for information exchange between all-comers will be lost!

  • Counterpoint: Absolutely not! Quite the opposite. Suppose that the network is supported only by connection fees. The connection fee will then be set based on the average usage for a connection of a given size. Then the small users will be paying more than their share to support the heavy users. A corollary to this myth is that the user cost of the Internet will increase if responsive pricing is introduced. The whole purpose of responsive pricing is to make the network more efficient, and to raise the value for users. Thus, if implemented intelligently, we will get more value out of a network of given cost. For a network of fixed size, we can lower the connection fees by an amount equal to the congestion fees and still recover costs, so total outlays are the same but the network has higher value. Or we could use the congestion revenues to invest in a bigger, more valuable network facility.

    It is also worth remembering that with responsive pricing, you pay for your actual usage in terms of the cost it imposes on other users. If all you want to use the Internet for is email and netnews, your charges under responsive pricing would be zero because these are flexible applications and do not require real-time performance or guaranteed bandwidth. As for rich users being able to afford to ignore dynamic prices, this is true under any pricing scheme and is a larger issue concerning the distribution of wealth in a society.[g]

• Responsive pricing is just another way for network operators to make more money. Users will lose out as network operators maximize their profits.

  • Counterpoint: It's true that there is the potential for profiteering whenever prices are charged, especially when the conditions under which prices are set are not immediately accessible to ordinary users. But in a competitive environment, the market disciplines network operators whose revenues exceed actual cost by more than the minimal amount necessary to stay in business. Of course, market discipline is limited in the case of a monopoly provider or a cartel of price-fixing providers. But then the outcome depends on policy and regulatory decisions rather than on the specific pricing scheme.

• By introducing responsive pricing, the traditional Internet culture (which emphasizes openness and sharing) will be destroyed, and it will become just another commercial service.

  • Counterpoint: It's true that by changing the pricing scheme used in the Internet, the culture will also change. However the culture is changing anyway due to the strains imposed by the demands of the ever-increasing user population. The question is, how can this change be managed so that overall user satisfaction with the Internet is maximized? Insisting that the Internet remain a connection-fee only network consigns it to ever-lessening value as usage and congestion increases, and new, QOS-sensitive applications are developed that cannot be successfully implemented in a first-come-first-served network. Many users are already fleeing the increasingly noisy Internet; shifting some responsibility for congestion control out to the users, and treating them as smart rather than dumb devices, will help preserve those parts of the Internet that users most value preserving.

• We don't know what the future Internet will look like, so it would be a mistake to adopt a responsive pricing scheme which is so controversial — it could stifle innovation and cause the Internet to miss out on opportunities to enhance its value to society.

  • Counterpoint: Equally, by not introducing some additional forms of congestion control, the Internet may miss out on future growth and improvements. For example, the Internet may be consigned to missing the widespread deployment of real-time interactive video if better mechanisms for controlling congestion are not developed. We propose one particular form of congestion control based on economic principles of pricing for resource allocation. Price is one possible feedback signal which has some attractive properties (compactness, quantifiable, etc.). Economists have developed a large body of theory of pricing mechanisms, and there is a lot of experience with the use of prices in real-world markets. However we do not rule out the possibility that there are other feedback mechanisms that, for one reason or another, may be preferable in communication networks.