Prefetching data cache based on SSD device in distributed file systems

Abstract

A storage device that is used mainly in the Hadoop system, HDDs(Hard Drives Disks) have slower transfer speed and higher latency than SSDs(Solid State Disks). Even though the SSDs as an alternative to HDDs will replace HDDs eventually, some distributed file systems used to use SSDs as a cache of HDDs because SSDs are still too expensive. In this paper, we propose an efficient prefetching mechanism for distributed file systems that use SSDs as a cache. The proposed mechanism acquires a block ID that already used or will be used and prefetching it into the SSDs cache before receiving I/O requests. The proposed mechanism can increase the SSDs cache hit ratio when Hadoop MapReduce tasks are executing. In our performance evaluation, results show that the proposed mechanism significantly increased the performance of the Mapreduce compared with existing distributed file system where the existing SSDs cache was applied. © 2016 International Information Institute.