Concurrent diagnosis in parallel systems

The problem is considered of designing homogeneous multiprocessor systems in which concurrent 1-step diagnosis is possible. For systems whose interconnection structures are defined by regular graphs (called D//1//L systems), lower bounds and upper bounds on the maximum number of modules involved in computation (degree of parallelism) have been derived and strategies for the assignment of computational tasks to modules have been given. Optimal interconnection structures which enable concurrent 1-step t-fault diagnosability have also been presented. Some more work is needed to take into account the fact that in general required connections between busy modules may be constrained by the algorithms to be executed in the system. A more integrated study of the constraints and the tradeoffs required by computation and diagnosis will be needed in order to more fully exploit the potential performance improvement of multiprocessor systems.