A moving average method for predicting process resource usage based on a state transition model

We develop a prediction algorithm for process resource usage based on a state transition model. The state transition model is built by using a k-means clustering algorithm applied to a series of 2-dimensional observed parameters on process resource usage such as load average and free memory in a computer. Our prediction algorithm estimates the parameters from state transition probabilities of the model. To reduce prediction error, we introduce a moving average method in the prediction algorithm.