The Parallel Virtual File System for Commodity Clusters

One benefit of cluster computer architectures is the opportunity for large I/O bandwidths. High performance applications that require significant I/O throughput are increasingly of interest to the clustercomputing community. Parallel file systems are critical system software components that allow parallel applications to take advantage of the parallel I/O disk subsystems in a cluster architecture. The Parallel Virtual File System (PVFS) was the first parallel file system developed specifically for cluster architectures and today is a popular choice for fast disk access. This tutorial introduces the basic concepts behind parallel file systems and specific information about PVFS. The emphasis is on using PVFS to implement effective applications that require substantial I/O performance. Topics include fundamental concepts, PVFS specifics, I/O interfaces including MPI-IO, example applications, performance tuning, and the future of PVFS. Dr. Ligon is an Associate Professor in the Holcombe Department of Electrical and Computer Engineering at Clemson University. His research interests are in high performance computing, focusing on parallel file systems, problem solving environments, and reconfigurable computing. Current projects include the Parallel Virtual File System (PVFS), Clemson Environment for Remote Sensing (CERSe), and the Reconfigurable Computing Application Development Environment (RCADE). Dr. Ross is a postdoctoral researcher in the Mathematics and Computer Science Division at Argonne National Laboratory. His research interests are in message passing and storage systems for high performance computing environments, in particular cluster computing environments. Current projects include the Parallel Virtual File System (PVFS), the ROMIO MPI-IO implementation, and the MPICH implementation of the MPI message passing interface. Proceedings of the 2001 IEEE International Conference on Cluster Computing (CLUSTER’01) 0-7695-1116-3/02 $17.00 © 2002 IEEE