A partitioning method for efficient system-level diagnosis

We propose a partitioning method for an adaptive distributed system-level diagnosis in arbitrary network topologies. It utilizes a biconnected component as a partitioning unit. In an adaptive distributed system-level diagnosis, testing assignment algorithm is performed before each node performs actual diagnosis to reduce the number of tests in the system. Existing testing assignment algorithms adopt non-partitioning approach covering the whole system, so they incur unnecessary extra message traffic and time. In our method, the whole system is partitioned into small groups (biconnected components), and testing assignment is performed within each group. By exploiting the property of an articulation point of a biconnected component, initial testing assignment of our method performs better than non-partitioning approach by reducing the number of nodes involved in testing assignment. It also localizes the testing reassignment caused by system reconfiguration within the related biconnected components only. It is shown that our system-level diagnosis method is correct, and the number of messages required for testing assignment of our method is smaller than that of the previous non-partitioning methods. Comparisons with other partitioning approach are also presented.