Internet Cost Allocation and Pricing

EN: This article was written when the Internet was going through the transition from a partially subsidized public network to a fully commercial one - and as such it defines theories for a future that has already come to pass.

There is some concern that the transition from a government-owned network to a public one with private ownership will be damaging to the Internet itself. The author undertakes to apply economic theory, to derive a possible future that is based on free market principles.

Quantity of Internet Service

Under central control, the Internet's service model is "best effort" - the network allocates bandwidth among all the users in the same instant as best it can, and attempts to serve them without an explicit commitment to service quality, and the users have no recourse against the network in instances where the quality of service results in failure.

As the Internet population increases, and additional media (audiovisual, notably) are provided, the total data transfer is certain to continue to increase, and congestion will become a serious issue. As a public good with a best-effort service model, users have no basis for complaining, and no-one to whom they may complain even if they did.

The concern is that, without regulation, congestion will continue to increase, to the point that users will attempt to "game" the system to get more than their fair share of bandwidth. That is not to say that regulation would be much better: especially in the case of government regulation, priority would be given to certain users (chiefly the defense agency and research labs that began the Internet), and the remaining bandwidth would preclude other uses.

Free-market economics provides a solution to the concern about quantity: the current resources would be allocated to the users who need them most, as evidenced by their willingness to pay a premium for their traffic to be given priority.

Where demand exceeds supply, new suppliers are drawn to the market to capitalize on the opportunity to serve customers at lower price points, until the market finds an equilibrium where buyers receive the full quantity they are willing to purchase, at a price they are willing to pay for it.

Further, as with material goods, there will be a market for the bargain-seeker (who is willing to sacrifice quality/reliability in exchange for a lower price) and the high-end consumer (who is willing to pay a premium for highly reliable service) - and there will be suppliers to serve their needs.

Price of Internet Service

Do date, there have been two models for pricing Internet service: a flat-rate for a specific connection speed, or a usage-sensitive rate that charges per unit of data transferred.

To date, flat rates have been preferred by customers on various levels, as a predictable fee reduces their risk (or getting a higher than expected bill, or being "cut off" unexpectedly), but there are some instances in which usage-sensitive rates are more economical (the customer who receives large amounts of data periodically, but has no need for access in between those times).

The author explores specific instances in which one of the models is superior or inferior to the other, but ultimately it will be in the nature of demand to work out the best solution: billing structure is a product feature that itself is subject to supply and demand.

A Specific Proposal for an Expected Capacity Service

The author proposes a service model designed to avoid packet loss, in that when congestion occurs on the network, packets are queued - and rather than being dropped when capacity is insufficient, they would merely be flagged for later delivery. From an economic perspective, it seems to make sense - but from a technical perspective, it's completely wrong-headed.

While queuing packets for later delivery would help bandwidth, it ignores the fact that the queue would need to be stored somewhere, so instead of a bandwidth logjam, there would be a storage overload - and when storage capacity is exceeded, packets will be dropped anyway.

Also, certain goods have time sensitivity - a patient who needs a doctor immediately may not need one the next week (as he'll be dead), and information that is valuable now (the current weather conditions, departure time of a flight, etc.) may be worthless a few hours in the future.

The Heterogeneous Internet

Another problem with bottlenecks is the misidentification of the source: since a packet flows across the wires of many providers, the customer's frustration at the quality of service may be directed at his own service provider, for a problem that was on another company's wires.

The practical solution to this dilemma has largely been addressed by the earlier chapter on network interconnection. The economic solution is that a service provider, to be able to offer a "quality' of service its customers expect, must carefully manage its interconnection agreements and build out its own networks where such agreements fail to meet its quality criteria.

In that way, the quality of an internet connection will be related to the price its customers pay, and there should arise suppliers at various price points to appease the reasonable demands of customers.

Combining Sender and Receiver Payments

The author suggests that payments for use of bandwidth can be split between sender and receiver - but my sense is that this is not a practical suggestion: if a receiver is charged for bandwidth, there is the potential for his bill to be affected by the actions of others (per the problem with inbound e-mail in the new Zealand example).

The author further suggests a solution where a sender might "pay" for a packet's transmission to a certain point (MAE East), and the receiver would then be required to pay for the rest of the journey.

While theoretically possible, this is also highly improbable. To draw an analogy, it would be like having a postal system in which the recipient pays the postage, or part of the postage, for items sent him by others.

However, this plays out in reality to some degree, in the topography of networks: an ISP can arrange for a connection from their NOC to a MAE, such that they pay for the cost of the network to that point, but packets must be carried on other wires after that point. The cost is not, however, apportioned directly to a specific recipient for the transmission of a discreet packet, but is instead borne by their ISP as part of an aggregate total.

Static vs. Dynamic Payment Schemes

The author continues on his goose chase, theorizing that individual data packets could be tagged, enabling the sender to indicate the destination and how much of the postage he is willing to pay, while the received indicates the senders he is willing to receive from and the postage he is willing to pay.

In instances where the conditions set by both sender and receiver are met, the transmission would take place and payment would be automatic. In instances where the conditions are not met, the receiver may examine the tags of the various packets addressed to him and decide, dynamically, which to accept.

This, too, is a plausible theory that is not practical to implement: it would increase the needs for storage (information about "tags" would need to be stored at every router in-between), and my sense is that the traffic generated by messages about tags could well exceed the traffic to send the data to which the tags pertain.


The author concludes that, when traffic becomes a problem, solutions such as the ones he has described could be implemented to alleviate it.

EN: In all, not a very useful article.