Reliability Analysis of Energy-Efficient Parallel Storage Systems

Existing reliability models evaluate the lifetime of storage systems well. However, few models aim to analysis parallel storage systems, especially systems equipped with energy-efficient solutions. MREED model provides an inspiring reliability analysis idea for energy-efficient parallel storage systems, especially RAIDs. Unfortunately, the MREED has its disadvantage that it can evaluate RAID-0 only, in which a basic method of data-stripping is applied. In this paper, we extend the MREED model to cover the most energy-efficient RAID setups including RAID-1, which introduces disk-mirroring techniques, and RAID-5, which applies a data-parity techniques based on data-stripping schemes for a data recovery purpose. The MREED is then applied to evaluate an existing well-known power-aware RAID system-PARAID. Experimental results indicate the data locality affection and data parity impaction on the reliability of PARAID. Furthermore, we investigate a fundamental trade-off between reliability and energy-efficiency in terms of energy-saving RAID systems. A tradeoff curve is provided to justify whether it worth to sacrifice reliability for better energy efficiency.