The paper proposes the application of modern control-theoretic techniques like stochastic filtering, estimation, prediction, and time series analysis, etc., for performance optimization. It is argued that the queueing theory, which has been the most commonly used tool for computer systems modeling and performance studies, is limited in its scope due to its static nature. Further improvement in performance can be obtained by dynamic optimization. Therefore, control theory provides a promising future direction for performance improvement. A general methodology for formulation of operating systems resource management policies using this approach is presented. Some of the results obtained using this methodology in the area of CPU and memory management are also briefly discussed.
Complete paper in Adobe Acrobat format.