Reliability Evaluation of Clustered Faults for Regular Networks Under the Probabilistic Diagnosis Model.

This article focuses on extending and analyzing a probabilistic fault diagnosis algorithm for large-scale multiprocessor systems modeled as regular networks. It generalizes the diagnosis threshold from \( t \leq 2 \) to \( t = 3 \), enabling more accurate identification of faulty and fault-free processors by examining clusters of nodes called factions. The study provides detailed local and global performance analyses of the algorithm, employing Poisson and Binomial distributions to evaluate diagnostic accuracy, and demonstrates that the method achieves high correctness as network regularity (degree \( k \)) increases. The approach is applied to well-known regular network topologies—complete cubic networks, dual cubes, and hierarchical hypercubes—highlighting its theoretical foundation for assessing reliability in these systems. The paper also notes future work will address probabilistic diagnosis in irregular network structures.