Analysis of some algorithms for use on paged virtual memory computers

Abstract

Handling a single page fault involves execution of thousands of instructions, drum rotational delay and is usually so expensive that if it can be avoided, almost any other cost can be tolerated. Optimizing operating system performance is usually the main concern of computer seientists who deal with paged memories. However, redesigning the algorithm used by a problem program can often result in a very significant reduction in paging, and hence in program execution time. The redesigned algorithm frequently does not satisfy the more conventional efficiency criteria. A sorting algorithm, Hash Coding and other search algorithms are considered. Analytic and simulation studies are presented, and aome modifications are proposed to reduce the number of page faults produced by data set references. Analysis is in terms of three of the most commonly used page replacement algorithms i.e. least recently used, first in first out, and random selection. The modifications are for the most part relatively minor and in some cases have appeared elsewhere in the context of searching on external storage media. The important aspects are the dramatic performance improvements which are possible, and the fact that classical internal algorithms are inappropriate for use in a paged virtual memory system.