A multiple subsystem simulator of processor scheduling

Abstract

A multiple subsystem simulator of processor scheduling for performance of a time-sharing system has been developed and validated. This FB₃ multiple subsystem utilizes an objective function which considers both processor time as well as memory requests. A previously developed iterative optimization procedure is extended to provide measures of performance optimization, i.e. optimization with respect to variables of the software and management control in a hardware fixed environment. This iterative optimization technique utilizing both an analytic and simulation approach applies to computer system models driven by either a general system load or an objective function based on general forms of cost of delay functions. The job load is described by two probability density functions; one which describes the interarrival time between requests and one which describes a request's need for processor time. The optimization procedure requires no assumptions concerning the description of these probability density functions. In general, either one or both functions can be described by histograms. The other component of the system load, management control, involves the choice of a family of Greenberger cost of delay functions. The optimization procedure requires no assumptions concerning the functional form of these cost of delay functions. The developed simulator is applied to several scheduler models from this class of optimization models, these models are solved, and the results validated. Representative models that are analyzed include Hellerman's scheduling functions, round-robin, FB₂ and FB₃. A mathematical multiple subsystem model, based on the work of Blatny, Clark, and Rourke, of processor scheduling for performance of a time-sharing system is constructed. The optimization procedure used to find the optimal cost of the FB₃ multiple subsystem is an extension from two to three-dimensional state space. These models were validated by comparing derived data to acceptable values from other models specified in the literature.