Finding Optimal Demand Paging Algorithms

Abstract

A cost is defined for demand paging algorithms with respect to a formal stochastic model of program behavior. This cost is shown to exist under rather general assumptions, and a computational procedure is given which makes it possible to determine the optimal cost and optimal policy for moderate size programs, when the formal model is known and not time dependent. In this latter case it is shown that these computational procedures may be extended to larger programs to obtain arbitrarily close approximations to their optimal policies. In previous models either unwarranted information is assumed beyond the formal model, or the complete stochastic nature of the model is not taken into account.

Additional details