Home-based Cooperative Cache for parallel I/O applications

Abstract

Recently, there has been much research about cluster computing to get high performance using low-cost nodes connected with high-speed networks. In many research fields, many I/O subsystems which access disks slower than any other component in cluster computing have been proposed such as parallel distributed file systems and Cooperative Cache. In many I/O subsystems, Cooperative Cache was proposed to reduce servers’ load and to achieve high performance by sharing nodes’ caches. Previous researches about Cooperative Cache focus on how to efficiently manage read caches and do not care about the write policy. In parallel I/O applications, which use cluster computing, there are many write accesses as well as read accesses. To improve the overall I/O performance, it is necessary to improve write performance as well as read performance. In this paper, we propose the home-based Cooperative Cache. In the home-based Cooperative Cache, every block has its static home. The home node of a block does the read caching and the write buffering of the block. We apply the home-based Cooperative Cache to Parallel Virtual File System (PVFS) and Cooperative Cluster File System (CCFS). Also, we evaluate and analyze the performance of the home-based Cooperative Cache in PVFS and in CCFS.