Abstract
Recently, energy consumption of IT devices including storage systems is extraordinarily increasing, and it has become a serious problem. In this paper, we propose a tiered storage method with an access prediction for saving energy consumption of high-speed mass storages in HPC systems. A high-speed mass storage with our proposed method consists of high-speed online storages (OL) and large nearline storages (NL). All data are always stored in NL. Before job execution, only the files accessed by HPC Systems are migrated from NL to OL. In order to save energy consumption, NL disks are usually powered off and on only while data migration. Our method predicts when and which job is executed and files are accessed based on job scheduling to manage data migration.
