International Journal of Computer Science and Technology

International Journal of Computer Science and Technology | Vol. 6, No. 8, August 2004 | pp. 57–64

DOI: 10.46882/2004/IJCST/000061

Article Type: Research Article

Title: Design and Testing of a Fault-Tolerant Distributed Shared Memory Protocol for Linux Clusters

Names of Authors: Sophie Dubois¹, Hans-Werner Schmidt²

Authors’ Affiliations: ¹Department of Applied Cryptography, ENS Paris, France; ²Fraunhofer Institute for Secure Information Technology, Darmstadt, Germany

Abstract: Distributed Shared Memory (DSM) models simplify parallel programming pipelines but are prone to complete cluster failure if any individual computing node goes offline. This study outlines a robust, page-based Software DSM protocol that incorporates an asynchronous checkpointing infrastructure to guarantee fault tolerance. The software architecture tracks page modification actions using a relaxed, lazy release consistency model, saving differential page updates to non-volatile local storage blocks. Fault recovery evaluations conducted on a 32-node Linux workstation grid showed that our protocol handles unexpected node terminations cleanly, restoring cluster processing state within 1.45 s of failure detection. The performance penalty introduced by the checkpointing code loop caused a negligible throughput drop of less than 2.8% under typical matrix factorization workloads, offering a secure environment for long-running industrial compute loops.

Keywords: Distributed Shared Memory, Fault Tolerance, Workstation Clusters, Checkpointing, Consistency Models

Manuscript Timeline: Received: May 14, 2004; Revised: June 22, 2004; Accepted: July 19, 2004; Published: August 15, 2004

Citation: Dubois, S., & Schmidt, H.-W. (2004). Design and Testing of a Fault-Tolerant Distributed Shared Memory Protocol for Linux Clusters. International Journal of Computer Science and Technology, 6(8), 57–64.

Subscribe to read the full article: https://internationalscholarsjournals.org/subscribe-to-read