Issue No. 07 - July (2002 vol. 51)
<p>The problem of identifying the faulty units in regularly interconnected systems is addressed. The diagnosis is based on mutual tests of units, which are adjacent in the "system graph" describing the interconnection structure. This paper evaluates an algorithm named EDARS (Efficient Diagnosis Algorithm for Regular Structures). The diagnosis provided by this algorithm is provably correct and almost complete with high probability. Diagnosis correctness is guaranteed if the cardinality of the actual fault set is below a "syndrome-dependent bound," asserted by the algorithm itself along with the diagnosis. Evaluation of EDARS relies upon extensive simulation which covered grids, hypercubes, and cube-connected cycles (CCC). Simulation experiments showed that the degree of the system graph has a strong impact over diagnosis completeness and affects the "syndrome-dependent bound," ensuring correctness. Furthermore, a comparative analysis of the performance of EDARS, with hypercubes and CCCs on one side and grids of the same size and degree on the other side, showed that diameter and bisection width of the system graph also influence the diagnosis correctness and completeness.</p>
Fault tolerance, system-level diagnosis, PMC model, multiprocessor systems, wafer-scale testing.
A. Caruso, S. Chessa, P. Maestrini and P. Santi, "Evaluation of a Diagnosis Algorithm for Regular Structures," in IEEE Transactions on Computers, vol. 51, no. , pp. 850-865, 2002.