コンピュータシステムのスケジューリングと処理効率 : 数理と非数理の狭間の一局面

抄録

Empirical studies suggest that scheduling policies for CPU processing and job loading affect the processing capacity of computer systems. These policies use the characteristics of how each job requests hardware resources. This review presents, firstly, the process of how the author has been studying the above scheduling problem. The difficulty of the problem lies in the fact that the general model to which the practical engineers want to have solutions is analytically intractable. Secondly, this review presents some properties of scheduling on the basis of the theoretical results on restrictive but analytically tractable models that may give us a solid understanding of the effect of scheduling on the processing capacity of computer systems. These studies show that the simple CPU scheduling policy whereby I/O bound jobs have preemptive priority over CPU bound jobs is effective and shows sound behavior in all cases. On the other hand, the optimal job loading policy is complicated and its unsuccessful implementation may easily lead to serious performance degradation.