Pricing Interactive Computer Services
W. A. Gale · The Computer Journal · 1984
This paper illustrates a method for setting theoretically and empirically defensible prices for interactive computer services. The theoretical rationale is that prices should reflect the marginal social cost for using interactive computing measured by the delay imposed on other users. The empirical basis for the prices is the direct measurement of delay as a function of load, and of average load levels as a function of time. An intrinsic problem of shared resources such as interactive computer systems is congestion. Every user contributes to some degree to the general deterioration of service quality (response time) as system loads increase. Since capital and operating costs of computer systems are fixed irrespective of system load, the delay costs imposed on users by congestion are the predominant marginal costs of system operation. It is an economic theorem that unless prices are set to marginal social costs, inefficient use of resources results. Therefore, efficient use of interactive computing services requires setting prices to equal the expected cost of marginal congestion delay. The paper shows that under plausible conditions, marginal congestion prices will cover costs of operation. Furthermore, profitability with such prices in effect gives information on when to make additional investment in capacity. We also show how to construct statistical models of response time from direct observations. Direct observations can provide a robust and understandable empirical basis for marginal congestion pricing. The empirical work needed is illustated by measurements on a VAX 11/780 offering UNIXTM service. We show the use of several statistical techniques to display the results understandably.