Evaluating disk idle behavior by leveraging disk schedulers

Abstract

Disk idle behavior has a significant impact on the energy efficiency of disk storage systems. For example, accurately predicting or extending the idle length experienced by disks can generate more potential opportunities to save energy. This paper employs a trace driven simulation to evaluate the impacts of different disk schedulers and queue length thresholds on the disk idle behavior. Experimental results give three implications: (1) Position based schedulers and long queue length thresholds can significantly reduce the maximal queue length and the average queue length. (2) Position based schedulers and long queue length thresholds can generate more idle periods which are shorter than 1 s, but they do not affect those long idle periods contained in the modern server workloads. (3) Disk idle periods demonstrate both self-similarity and weak long-range dependence, and the disk schedulers and queue length thresholds do impact the Hurst parameter and the correlation behavior of the workloads. The analysis results in this paper provide useful insights for designing and implementing energy efficient policies for the disk drive based storage systems.