Statistical and Pattern Based Models for CPU Burst Prediction
G. L. Lasseter, Lo, T., K. Mani Chandy, James C. Browne · 1973
CPU burst durations for several jobs were measured. The CPU bursts exhibited pat- terns over time. A partially observable Markov process was used to model CPU burst generation. The model seems to fit data reasonably well. A simulation study on general prediction schemes was carried out. A job's predicted burst was made to fall within a confidence region with a certain probability. The effect of the length of the confidence region and the probability of predicting within that region were studied. The device utilizations were remarkably insensitive to predictive accuracy. A pattern based predictor was developed and it yielded satisfactory schedules.