A tool for detecting transient faults in execution of parallel scientific applications on multicore clusters

• Published in: Journal of Computer Science & Technology Vol. 14; no. 1; pp. 32 - 38
• Main Authors: Montezanti, Diego, Rucci, Enzo, Rexachs, Dolores, Luque, Emilio, Naiouf, Marcelo, Giusti, Armando De
• Format: Journal Article
• Language: English; Spanish
• Published: La Plata Graduate Network of Argentine Universities with Computer Science Schools (RedUNCI) 01.04.2014; Universidad Nacional de la Plata, Journal of Computer Science and Technology; Postgraduate Office, School of Computer Science, Universidad Nacional de La Plata
• Subjects: Fault detection; message content validation; Middleware; Multiprocessing; parallel scientific application; soft error detection tool; transient fault
• ISSN: 1666-6046; 1666-6038

Transient faults are becoming a critical concern among current trends of design of general-purpose multiprocessors. Because of their capability to corrupt programs outputs, their impact gains importance when considering long duration, parallel scientific applications, due to the high cost of re-launching execution from the beginning in case of incorrect results. This paper introduces SMCV tool which improves reliability for high-performance systems. SMCV replicates application processes and validates the contents of the messages to be sent, preventing the propagation of errors to other processes and restricting detection latency and notification. To assess its utility, the overhead of SMCV tool is evaluated with three computationallyintensive, representative parallel scientific applications. The obtained results demonstrate the efficiency of SMCV tool to detect transient faults occurrences. Keywords: Transient fault, parallel scientific application, soft error detection tool, message content validation.