Economic Considerations

A major selling point for the virtual circuit based network technologies is that those technologies are able to collect per-call billing information, allowing providers to charge customers for usage, with distance-sensitive rates.

A number of opinions appeared recently in various publications claiming (in different combinations) that:

  • proponents of flat rate are in fact for free-for-all connectivity,
  • flat rate does not create incentive for backbone service providers to improve their service,
  • flat rate does not generate enough revenue to sustain growth of the Internet,
  • flat rate stimulates irresponsible usage, thus creating congestion,
  • there is a pressing need for charging schemes similar to POTS charging of call initiator and to reverse charging, to remove the burden of paying for traffic from the content providers,
  • real-time applications need to be billed separately, because they harm asynchronous traffic,
  • congestion-based pricing is desirable to incent ISPs to increase capacity,
  • per-connection billing is feasible technically and economically,
  • there's no alternative to virtual connections or resource reservation to achieve differentiation of quality of service offerings.

The first claim is obviously more a political rhetoric than anything approximating the truth. Apparently that point of view is a reaction on the enthusiastic statements that the "bandwidth is too cheap to meter" taken out of context. The current flat-rate Internet infrastructure is definitely not free, and users pay for the connectivity either directly or in form of government or corporate investments in the infrastructure. The flat-rate usage is also not "unlimited", despite what many sales pitches tell. The traffic of every customer is limited by the capacity of his customer-access facilities, and the connectivity price universally depends on the customer access line capacity.

Part of the revenue generated by the flat-rate, small service providers goes to backbone service providers, who spend it on transportation of data planet-wide. Obviously, the economical interest of a large ISP is to spend as little of possible on the global transport infrastructure and charge customers as much as possible. The product they sell has two parts: first is the actual delivery of data, and the second is the potential to deliver data everywhere. (The very success of the Internet is confirmation that customers do value the potential connectivity, even if they use only a very limited subset of it.)

With the actual delivery the situation is simple, and the delivery quality is measurable by relatively straightforward methods. In a sense, the small ISPs act as wholesale brokers on behalf of their clients, trying to find the overall best deal on quality of service and prices of a backbone provider. This places backbone providers in direct competition with each other, thus the market pressure forces them to offer reasonable service at reasonable prices. In fact, the power of that pressure was exemplified by the meteoric entrance of Sprint into the backbone ISP market, when many small ISPs found the absence of "reseller" surcharges and the DS-3 backbone appealing and thus placed Sprint in first place market share-wise in a little more than a year.

The economics of potential connectivity is more interesting, as it places backbone service providers in a situation where they are forced to cooperate. Essentially, a backbone provider that does not provide "universal" connectivity has zero chance of selling its service to general Internet customers, and so will only be useful as a carrier for corporate private networks. The universal connectivity requirement is in fact very costly because in necessitates deployment of backbone routing equipment which has to be able to work with full routing tables and perform all necessary routing policy computations, in addition to packet forwarding. It also makes public backbones much less stable because global networking tends to require leading-edge equipment and software, i.e. technology that is not well-tested, expensive, generally unreliable, and requires highly qualified staff to operate and maintain.

The same market pressures make all the incentive the backbone Internet service providers need to keep on improving service or offer better pricing. Customer mobility is relatively high in the market, and there is little (if any) brand loyalty; the services rendered have no distinctive color, taste, or any other individually preferable feature; there is no compatibility problem; so different service offerings are simply of worse or better quality, at different prices. It means that a provider with an inferior service offering is quickly deserted by its customers. Therefore, the second claim is completely groundless.

The question of whether the revenue stream generated by the present flat-rate service is sufficient to keep the Internet afloat is also interesting. All large ISPs are essentially money sinks; they do not generate revenue to pay back the investments because they are forced to make massive investments into infrastructure just to keep on par with the competition. On the other hand, their value rises at unprecedented rates, as shown by the example of UUNET, which became a two-billion dollar corporation from a small business in just few years. In other words, ISP business is very much like the PC business when it had just started - there will be losers, but winners will make their shareholders rich. So, although Internet services do not generate revenue to pay off the investments, they do produce very rapid growth of capital, i.e. the scenario of the Internet collapsing because of an extreme shortage of funding is unlikely. The situation will change with introduction of the new generation of scalable backbone routing equipment (such as that of Pluris, Inc.), which would allow the amortization of backbone equipment over 10-15 years instead of just 3-4 years (the current period of obsolescence of existing backbone routers).

The claim that flat rate promotes irresponsible usage (so economical pressure is necessary to keep the usage down) does not seem to represent reality: the time people spend on-line does not depend on the performance of the network (as long as it is not so poor so as to border on uselessness), or even decreases with improved performance (less waiting). The likely explanation is that people do not indulge in usage for the sake of usage, rather they want to reach some specific places and when the task is accomplished, log out. Therefore, any imposed usage-based charging would not cut "unproductive" usage, but rather reduce the utility of the network to its customers. It is hardly a solid foundation for the sustainable growth.

Many proponents of per-connection billing are pointing out that content providers must somehow recover their communication costs and that it would be a good idea to pass those costs on to users. This argument ignores the fact that communication costs represent only a small fraction of costs necessary to create and host original content; so if communication structure is geared to recover costs, it has to be able to pay back content providers. The historical precedent for this type of service (900 numbers), however, shows that most businesses are interested in more than recovery of costs and need more flexible pricing plans for their services (in most cases, the value of service has nothing to do with the amount of data transferred). The result is that only those businesses that are not interested in any meaningful relationship with customers make use of the telco-provided collection, leaving 900 numbers a province of sleaze and astrology.

It is ironical that the recovery of transport costs becomes a pressing issue for content providers only when their usage is metered. There is no reason to think that the absence of network-based recovery of transport costs harms content providers in any way. In fact, free content providers are directly interested in publishing free "infomercials" (as most corporations do, to improve their customer relationship, reduce expenses on technical assistance, and attract new customers), are placing paid advertisements, are funded by public agencies (government and educational sites), or simply want to spend some (small) amount of money for vanity. There are a number of successful businesses providing information by subscription only, and when digital payment technology gains wider acceptance they won't even need to ask credit card numbers or learn identities of their customers.

An often repeated statement is that although flat rate is sufficient for asynchronous applications the real-time applications need usage-dependent pricing because they would create congestion. The real difference is not between asynchronous and real-time traffic, but rather between well-behaved and broken-as-designed real-time communication protocols. A variable-rate real-time communication protocol which has the same congestion control features as TCP does not create any additional congestion in backbones. In fact, it won't even require more bandwidth than a routine file transfer (which would take "infinite" bandwidth, if allowed). The current voice delivery technology allows telco-quality transmission voice in a bandwidth comparable to average traffic generated by a web browser.

Several economists pointed out that facilities are built to accommodate peak usage, so the prices must be increased during the periods of congestion to reduce usage during that time. This, however, completely ignores the unique property of TCP/IP, namely, its graceful degradation and fairness in regard to sharing bandwidth. That feature makes facilities to be loaded at close to 100% during practically all daylight hours, while the quality of service varies depending on the number of users. The non-locality of Internet traffic also tends to spread the load. Charging users more for lower quality service does not seem to be a particularly good policy; and in the absence of government regulation that would only cause providers to maximize revenues by creating congestion wherever possible. An access pricing dependent on time of day certainly can be implemented with off-the-shelf technology, but the author is not aware of any Internet service provider offering such pricing.

The technically feasible usage-based Internet pricing methods include accounting of connection time, and accounting of traffic over customer access circuits. Such pricing methods are widely used, and produce a single line on a customer bill (so there are no expensive itemized billing and associated billing disputes). However, the most discussed pricing method - the telco-style distance-sensitive (whatever the Internet distance is) per-connection billing requires connection-oriented transport, and an elaborate system of inter-provider settlements. Even if such a network was technologically feasible (it isn't, as was discussed in previous chapters), it is hardly economically feasible.

It is common knowledge that telcos spend a significant portion (from 15% to 70%, by different estimates) of their revenue not on providing actual service but on accounting, billing, collection and billing-related customer service. In the Internet, with its much lower bits/connection rate, the corresponding expenses would be in 95-99% range. In other words, a flat-rate service would be more than ten times cheaper than a distance-sensitive per-connection billing service for the same amount of information transferred, for an average user. That leaves a very small segment of the user population that would benefit from usage-based pricing; so the capital expenses necessary to implement such accounting and itemized billing would significantly outweigh any benefit an ISP would gain by attracting those customers.

The final argument pro virtual connection-based networks is that they provide customers with a variety of qualities of service. However, as we have shown in the previous chapter, the only type of service that cannot be approximated by a best-effort delivery network is support for isochronous (constant bit rate) traffic. Such traffic is purely an artifact of obsolete uncompressed voice transmission technology. It hardly makes any sense to cripple the new medium severely just to support a legacy application (alternative methods to support POTS traffic, such as edge translation and translating gateways can be used instead; an even simpler alternative is to leave POTS as it is).

The conclusion is, therefore, that there are no compelling economic reasons to prefer any virtual-connection based network technology over the existing best-effort delivery network.