Increasing data replication improves the reliability and availability of the large-scale storage systems. However, multiple replication required much more storage capacity and disk I/O frequency that cause of increasing the power consumption of the storage systems. To address this issue, we propose two data placement policies, Disk Group Aggregation and Cache Striping. These data placement policies employ different data mapping between buffers (memory) and disk drives to control buffer overflow timing of each replica to reduce the disk access frequency. In addition, we also propose two buffer flush algorithms, WithAllSpins and SpinupEE. WithAllSpins flushes buffered data to currently rotating disks, whereas SpinupEE forces disks to spin up based on the estimated energy efficiency, and writes buffered data to the disk to make the buffer space fresh. We evaluated the effectiveness of our proposals using a simulation program and demonstrated that they can reduce power consumption, even if the data are replicated multiply.
CITATION STYLE
Hikida, S., Le, H. H., & Yokota, H. (2019). Energy Efficient Data Placement and Buffer Management for Multiple Replication. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 11707 LNCS, pp. 19–29). Springer. https://doi.org/10.1007/978-3-030-27618-8_2
Mendeley helps you to discover research relevant for your work.