|
Abstract : |
This paper introduces a new fault model for systemlevel diagnosis and a class of on-line distributed diagnosis algorithms that operate correctly under the model. The algorithms are guaranteed to operate correctly in the presence of faulty nodes that disseminate arbitrarily corrupted diagnostic information. The fault model addresses the practical issue of designing an inter-node test to cover diagnosis algorithm operation. Since an explicit test to detect arbitrary failures is not practical, evidence of a node's faulty behavior is provided by examining diagnostic messages exchanged by the node. In many practical systems, algorithm overhead using the new fault model is only twice that required for algorithms using the PMC fault model. The key results of this paper include a description of the new fault model, the specification of a class of on-line distributed diagnosis algorithms that use this fault model, and proofs of their correctness., |
