Power Aware Disk Allocation

Disk power consumption is one of the major concerns in adopting applications with large scale I/O in both mobile and scientific computing domains. Virtualization and the resulting abstraction of large scale storage systems and variety in the I/O demands of applications call for a power efficient disk allocation strategy across simultaneously executing applications that provides necessary performance guarantees. In order to abstract the underlying diversity in capacities and rotation speeds of disks and attain a performance and power efficient allocation, a disk allocation algorithm has to often choose from a set of conflicting optimization criteria. This paper presents the trade-offs associated with power and performance across different disk allocation schemes, targeting a scenario where multiple applications exercise the same disk storage system at the same time. We also present a novel disk allocation scheme that reduces overall power consumption of a disk system while satisfying the performance and storage capacity constraints set by applications. Extensive analysis of our proposed disk allocation scheme shows that it reduces disk power consumption compared to other alternate disk allocation schemes, while providing similar or better performance guarantees.