Evaluation of design alternatives for a cluster file system

Abstract

Based on implementation experience and measurements, this paper presents an evaluation of design alternatives to a cluster file system. The file system is targeted for IBM cluster systems, Scalable POWER parallel and AIX HACMP/6000. We considered a shared disk approach where serialized, multiple instances of a single-system file system directly access file data as disk blocks, and a shared file system approach which is the conventional method of distributing file system function between a client and a server. We conclude that the shared disk approach suffers from the difficulties of metadata serialization, poor write-sharing performance, and read throughput.