Design and Implementation of Cooperative Cluster File System

摘要

This paper presents the design and implementation of Cooperative Cluster File System (CCFS), a parallel file system which supports users with efficient file I/Os in cluster computing. CCFS consists of three layers which are a single disk I/O, the home-based Cooperative Cache and a file system interface. The single disk I/O service was designed and implemented as a storage layer in CCFS. Single disk I/O services make disks distributed over nodes into a single virtual disk and provide users with high bandwidth by transparently stripping data over disks. The home-based Cooperative Cache as a file system cache layer in CCFS was used for the read caching and for the write buffering of data by sharing nodes' caches. These two main layers make CCFS more efficient to retrieve and to store data. We can reduce the management cost of metadata by dividing metadata updates into many small updates. As a result, CCFS showed better read/write response time than the previous cluster file system named Parallel Virtual File System (PVFS).