Performance evaluation of distributed diagnosis algorithms in parallel systems

摘要

This paper presents a quantitative analysis of distributed diagnosis algorithms. Using a process-based model, an evaluation tool called SimDiag that allows the evaluation of several diagnosis performance characteristics such as diagnosis latency, number of generated tests and transmitted messages, has been developed. Four well known algorithms which mainly differ in the way to establishing tests between processors are considered. the obtained results show the importance of using such tools to allow a distributed diagnosis algorithm to be efficiently used in realistic fault-tolerant parallel systems.