I.-C. Hwang, H. Jung, S.-R. Maeng, and J.-W. Cho (Korea)
Cooperative Cache, PVFS, Distributed File System, Cluster Computing
Recently, there has been much research to get high performance in cluster computing with inexpensive PCs connected by high-speed interconnection networks. One approach is to get high bandwidth and parallelism in file services by developing efficient distributed file systems. In various distributed file systems, PVFS (Parallel Virtual File System) provides users with high bandwidth by stripping file data over I/O servers in Linux. We developed cooperative cache for PVFS (Coopc-PVFS) because PVFS doesn't support any file system cache facility. Coopc-PVFS improves read performance by sharing clients' file system caches. However, it decreases write performance because it is a hint-based cooperative cache whenever a client writes a file, cache block invalidation messages must be propagated to all possible readers. In this paper, we present 2-phase protocol which can improve write performance by reducing write overhead in the original Coopc-PVFS. We show the efficiency of 2 phase protocol in comparison to the original Coopc-PVFS and PVFS. As a result, the response time of 2-phase protocol in Coopc-PVFS is shorter than or similar to that of the original Coopc-PVFS and that of PVFS.
Important Links:
Go Back
