Power-saving in storage systems for Internet hosting services with data access prediction

摘要

We present a power-saving method for storage systems in Internet hosting services, particularly those providing video/photo sharing services. The key idea behind our method is to skew the workload towards a subset of disks in the storage array, thereby extending the periods in standby mode of the other disks. Our method is based on the idea behind PDC, but the main objective of this study is to investigate a method that is adaptable to both constant massive influx of data and changes in data popularity over time. Moreover, to reduce accesses to disks in standby mode, our method periodically rearranges data in the order of potential future accesses presumed to be associated with elapsed time after upload and the accesses in the past, instead of just sorting according to the latest number of accesses. This correlation is obtained by analyzing access patterns for 45,000 randomly selected public photos on Flickr. Performance is evaluated through both simulation and a prototype implementation. In the experiments, we observed that our method saved 24.0% of running time of the disks in active mode, with an overall average response time of 47.6 ms, in which 0.43% of the total accesses involved disks in standby mode.