Simulation of a Model of Paging System Performance

by simulation is the performance of a probabilistic model of a multiprogrammed single-processor computing system operating under demand paging. Results of experiments on statistical methods .for improving the efJiciency of the simulation are presented. Estimates of the response variables in the simulation are reported for a variety of conditions of system overhead, queuing delays, and transient response. Sensitivity of these factors to the assumptions of the model are discussed. Many problems that arise in operating systems lead rather naturally to the study of queue networks. The models formulated, however, are frequently of greater complexity than can be handled by existing mathematical techniques. Queue network simulation by straightforward sampling is typically adopted, and often requires long simulation studies to adequately represent the behavior of queuing systems. It is advantageous, therefore, to investigate alternatives to straightforward sampling. Gaver' has proposed several such alternatives. Their value is suggested by data derived from a simple (though mathematically difficult) queuing situation. In this paper, we apply Gaver's methods to simulating the more complex queuing model of a paging machine previously described by Lewis and Shedler.2 Our approach compares three methods of simulation with straightforward sampling and then estimates the sensitivity of the system response variables to the assumptions of the model. In our discussion, we attempt to go beyond adescription of the experiment by illustrating a methodology for the simulation of probabilistic models. The computer system in our model is a multiprogrammed, single processor operating under demand paging. Overhead functions of the Central Processing Unit (Cpu) such as construction and execution of channel control programs, execution of replacement algorithms , and queue management are represented explicitly. Similarly,