Degree of multiprogramming in page-on-demand systems

Abstract

A simple stochastic model is described which offers a base for understanding the relationship between the number of programs permitted to share memory (the degree of multiprogramming), drum traffic rates, and central processing unit utilization in page-on-demand, multiprogrammed, time-shared computer systems. The model preserves, as a key feature, the property of page-demand statistics which implies a “burst” of page demands at the beginning of any job or quantum execution. The model, a Markov chain, is analyzed numerically and the results are presented graphically for a wide range of key environment-descriptive parameters. Implications of the results to time-shared system design and programming are discussed, and a calculation of the optimal degree of multiprogramming for a wide range of parameters is presented graphically.