Hirotoshi Akaike

Performance evaluation of energy-efficient high-speed tiered-storage system

Large-scale high-performance storage in high-performance computing systems consists of numerous HDDs and consumes a large amount of energy. To reduce the energy consumed by these storage systems, we previously proposed a high-speed tiered-storage system with a power-aware proactive method of storage-tiering management, which we named the energy-efficient high-speed tiered-storage system (eHiTS). eHiTS consists of a high-speed online storage component as the first tier and a large capacity nearline storage component as the second tier. The first-tier online storage is used as a data cache. For reducing energy consumption, it is necessary to reduce the online storage capacity, but this decreases performance. We derive the relationship between performance and online storage capacity by analytical calculation and simulation with an actual HPC system. The results show that with a 1% of decrease in performance, online capacity could be reduced to the capacity which can hold about 1.4 times as many volumes as the maximum number of volumes required by jobs running in parallel. In this case, energy consumption was reduced by an estimated 38% with eHiTS compared with conventional tiered-storage systems.