Transparency in a replicated network file system

Abstract

File servers acting as centralized data sharing and storage stations are very crucial in modern network environments. Failure and inefficiency of a file server would be unacceptable and thus data replication is necessary to provide fault-tolerant and high-performance services. In this article, we present a replicated network file service that follows the Sun NFS protocol. Replication transparency and failure transparency are emphasized in the design. Any NFS client implemented on existing systems can use our fault-tolerant service without any modifications. In the current prototype, the simple primary-backup model is used to ensure data consistency between servers. Two distributed algorithms were designed to monitor and maintain the server modes. Performance analysis shows that the prototype system is sufficient to support a mostly-read environment, and it is possible that a high-performance file service can be accomplish in future developments.