Efficient fault diagnosis using local inference


This paper proposes a Bay&an network approach to fault diagnosis in distributed computer systems which use test tmnsactions @robes). A lower bound on the MPE diagnosis accuracy is derived and analyzed with respect to the problem parameters. An empirical study of a localinference approximation scheme (algorithm approx-mpe( 1)) yields useful insights on the algorithm’s behavior and its applicability to the considered class of diagnosis problems: the approximation quality is higher for higher-MPE (“bigher-quality”) networks and “degrades gracefully” with noise.


    0 Figures and Tables

      Download Full PDF Version (Non-Commercial Use)