Seeking configurational optimization in computer systems

This paper develops analytic techniques for achieving macro-level configurational optimization of computer systems. The computer is viewed as a network of queues with the processors, main memory, auxiliary memory and the transfer channels being the servers. The objective function to be minimized is the total system cost divided by the probability of the CPU busy time for a given user benchmark. A queueing model is developed which evaluates this probability. The scope of optimization includes the determination of the parameters related to component selection and the determination of management oriented variables like degree of multiprogramming, main memory space allotted to each program. The optimization algorithm is a modified 'Direct Search Method'.