The Least Recently Used (LRU) method is the technique that is most often used for a page replacement algorithm. However, LRU may not be an effective method for some large‐array manipulation. A method for solving this problem is presented in this paper. In this method, the Operating System (OS) automatically detects loops that are not suitable for the LRU method. There is no necessity to modify application programs nor any additional overhead as long as sufficient real memory is provided. Therefore this method is most suitable for practical use. It functions on the basis of information about the relationship between the working set size and the window size, and indication of whether the detected page fault is caused by an instruction fetch or an operand fetch, etc. The results of experiments and improvements are also shown.

Queueing network models have established their role in performance evaluation of computer systems [1]. These models have a sound theoretical background, and require less computational effort than other techniques. And they have proved to be practical in many case studies. Evaluation using simple models is powerful, especially in predicting system behaviour, and makes the internal structure of computer systems more understandable than evaluation using benchmarks. However, the modeling approach forces us to validate the model. Queueing network models are no exception to this point.