A structural approximation method to generate the optimal auto-sleep schedule for computer systems

This paper addresses a problem of energy reduction on computer systems, namely the problem of determining the optimal auto-sleep schedule on which the system is changed into a low-power state. For this problem, we provide a stochastic model for a computer system with autosleep function. In our modeling framework, user requests arrive at the system in accordance with a general renewal process and are served by generally distributed service times. To obtain the optimal auto-sleep schedule minimizing the expected power consumption per unit time in the steady state, the phase-type approximation is applied to the renewal arrival process. In numerical examples, we investigate accuracy of the phase-type approximation through a simulation study.