Computing center optimization by a pricing-priority policy

Abstract

In this paper, a genera model for the optimal allocation of priorities through pricing is considered. The case of two priority queues ids discussed in detail. (for the m-priority queue analysis, the reader may refer to Reference 11.) In both cases, it is shown that a set of admission tolls can be established at the different priority queues. These tolls are based on user urgency, the job arrival rate, the expected service time, and the number of priority classes. By setting a different admission toll at each priority queue and by providing the user with information and motivation, he is encouraged to weigh the relative vlaues of the services before picking the priority for his job. According to his urgency, the user minimizes hsi total average cost function (cost to joint a certain priority plus the cost of delay). Pricing is addressed to the allocation of priorities in a computing system to minimize the total average cost of delay of an installation's user community. Each user is free to decide on the priority assigned to his job, and the optimality of the system is maintained. As a result, the computer can process first those jobs that are urgent and then proceed with less urgent jobs. The uncertainty of the cost per unit time delay c is considered in this model. It has been found, for the case of two priority queues, that if you have an estimate of the mean of the probability density function of the cost of delay per unit time ƒ(c), the cost differential (x 1 - x 2 ) is insensitive to the exact distribution. This result is feasible if we can redistribute the demand such that ρ ≤ 0.75. During periods with high traffic intensity it is shown that the admission tolls at the higher priority queues should be increased.