Predictive Power Conservation based on the Idle Time Pattern from Disk Access Data

We examined the power consumption of various dynamic disk spindown policies with the goal to save energy for mobile computing. Based on a trace from HP Laboratories, we find that the proposed adaptive 2-competitve algorithm can outperform the share algorithm by 15% or so when the disk spindown cost is low. The relative power performance of different disk spindown techniques are presented and compared. We focus on algorithms that explicitly exploit the burst nature of disk access pattern.