A scheduler based on the demand page‐stealing for large‐scale programs
Ken-ichiro Murakami, Takaaki Matsumoto, Masahiro Kurata, Masayoshi Nakao · Systems and Computers in Japan · 1986
Abstract Large‐scale programs (i.e., LSI‐DA) and simulation programs require a working set size of several Mbytes and a long execution time. When executed in parallel with interactive jobs, large‐scale jobs exert a degrading effect on the processing of interactive jobs. In addition, the turn‐around time for large‐scale jobs is degraded. It is therefore necessary to limit the amount of real memory which is made available for large‐scale jobs. However, a sound method for determining the limit has not yet been established. This paper proposes a method for improving the turn‐around time of large‐scale jobs without affecting the processing of interactive jobs executed in parallel. Turnaround time is improved by avoiding the swapping of large‐scale jobs. Furthermore, a look‐ahead push method is employed to ensure some given mean response time for interactive jobs when the memory load suddenly increases. Another feature of the method is that the limitation on real memory is adjusted automatically, thus eliminating the need for the user to set the limit. Simulation results indicate that the adverse effect on interactive jobs is minimized and that CPU utilization by large‐scale jobs is improved by a factor of two or more in systems in which a large number of terminals are connected.