Type of Culminating Activity

Graduate Student Project

Graduation Date

8-2006

Degree Title

Master of Science in Computer Science

Department

Computer Science

Major Advisor

Amit Jain

Abstract

The Beowulf Cluster Shell (beosh) has been created to provide cluster users with a Single System Image (SSI) by distributing individual commands or pipelined jobs over available nodes without requiring the user to be aware of where or how jobs are distributed. Job control of distributed jobs enables management of multiple concurrent remote jobs from a single shell session. In addition to SSI features, beosh supports parallel job execution through pdsh, a commonly used parallel-only cluster shell. In either distributed or parallel mode, beosh can limit nodes to those reserved through a system like the Portable Batch System (PBS).

This shell seeks to address the lack of a full-featured, portable cluster shell that remains despite increasing availability and use of dedicated, high-performance computer clusters. Many cluster \shells", for example, are actually scripts for local shells executing a sequence of remote commands for user convenience. At the other extreme are specialized cluster operating systems requiring commitment of the cluster to a particular operating system and its included utilities. Between these extremes, there are no choices for a general-purpose cluster shell comparable to even a basic single-system shell.

Though opportunities for improvement remain, beosh is already a usable shell providing a foundation for development of a full-featured, general-purpose shell for Beowulf clusters.

Download

Share

COinS