A fully automatic analytic approach to budget-constrained system upgrade

This thesis describes the development of a software package to upgrade computer systems. The package, named OPTIMAL, solves the following problem: given an existing computer system and its workload, a budget, and the costs and descriptions of available upgrade alternatives for devices in the system, what is the most cost-effective way of upgrading and tuning the system to produce the optimal system throughput? To enhance the practicality of OPTIMAL, the research followed two criteria: i) input required by OPTIMAL must be system and workload characteristics directly measurable from the system under consideration; ii) other than gathering the appropriate input data, the package must be completely automated and must not require any specialized knowledge in systems performance evaluation to interpret the results. The output of OPTIMAL consists of the optimal system throughput under the budget constraint, the workload and system configuration (or upgrade strategy) that provide such throughput, and the cost of the upgrade. Various optimization techniques, including saturation analysis and fine tuning, have been applied to enhance the performance of OPTIMAL.Science, Faculty ofComputer Science, Department ofGraduat