The allocation of resources in operating systems is currently based on ad hoc heuristics and algorithms. Some well-known heuristics have been developed in such areas as process scheduling, memory management and network routing. The possibility that better decisions could be made by an Intelligent Agent employing expert system techniques is investigated in this report. The ability to learn from experience and to deal with uncertainty are two characteristics of expert systems that would be essential aspects of such an Intelligent Agent.

As multiple processor systems become more widely available, applications involving multiple concurrent processes will increase in number and importance. This increased interdependency among processes poses interesting problems in the area of processor scheduling. How should the processes be scheduled to achieve some optimal level of performance? A scheduler based on an expert system may prove to be a viable alternative to those that have been proposed and (in some cases) implemented so far.

This report describes the implementation of a learning mechanism that attempts to handle the problem of processor scheduling in such a multiprocessor environment. In effect, the Intelligent Agent tries to "learn" its own set of heuristics for optimally scheduling a set of co-operating processes. By simulating a relatively simple multiprocessor system we examine the merits of such an approach.