| Summary: | This thesis applies optimal stochastic control theory, a well-known engineering technique, to adaptive performance control of computer systems. The major advantages of using the theory to adaptive performance control of computer systems are:
a) the policy so obtained can be shown mathematically to give optimal performance,
b) the policy is dynamic in the sense that control decisions are made based on the current system state thus optimizing performance at all time,
c) once a stochastic model of a system has been developed it can be used to determine optimal policies pertaining to different criteria simply by using different objective functions in the model.
The application is exemplified by using the theory to compute the optimal admission policy for a combined batch-interactive system with paging. A mathematical model of the system was developed and the theory applied to compute the optimal admission policy. The policy determines the number of batch and interactive jobs that should be activated at each system state. The solution is shown to exhibit the following properties:
i) It maximizes the total system throughput. ii) It gives good mean response time to interactive jobs.
iii) It guarantees a minimium level of batch throughput. The improvement in total system throughput compared to the
case when jobs are automatically admitted on arrival is substantial. The improvement is most dramatic when the workload is heavy. === Science, Faculty of === Computer Science, Department of === Graduate
|
|---|
