Optimizing energy and performance for server-class file system workloads
ACM Transactions on Storage (TOS)
Evaluating performance and energy in file system server workloads
FAST'10 Proceedings of the 8th USENIX conference on File and storage technologies
Sustainable predictive storage management: on-line grouping for energy and latency reduction
Proceedings of the 4th Annual International Conference on Systems and Storage
| Hi-index | 0.00 |
Recently energy has become a critical resource in modern computing systems. In mobile computers, disk sub-systems consume a major portion of available energy, which reduces battery life and thus limits the utility the whole system. In data intensive environments, the energy consumption of the I/O storage subsystem may easily surpass the energy consumed by the rest of the computing system. The direct and indirect cost of energy consumption is a limiting factor in deploying and operating data-intensive computing systems. This dissertation aims to provide energy saving solutions for both client side and server side environments, with the latter as the focus. For the client side systems, this work proposes a new energy-efficient file system called EEFS, to reduce energy consumption and improve performance by separately managing small-sized files with good group access locality. Because of the workload intensity, energy-efficient I/O solutions for client systems usually cannot be applied to data servers. Multi-speed disks were introduced by previous research work to make it possible to lower the disk energy consumption in I/O intensive environments. For multi-speed disk based server systems, this dissertation provides an energy-efficient RAID system named EERAID. Furthermore, this work addresses the open question of saving energy for conventional disk based storage servers by providing the policies EPRAID and eRAID, for RAID5 and RAID1 systems respectively.