System level diagnosis with local constraints

The concept of a local fault constraint in system-level diagnosis for multiprocessor systems is introduced. Given local constraints, distributed and sequential algorithms for a ring of processors as well as other regular interconnected structures are presented. In all cases considered, the maximum number of faults that can be diagnosed in the local domain is determined. The number of faulty processors that can be diagnosed when information about local constraints is available is significantly larger than what is allowed by the classical t-diagnosis theory. In addition, the number of permissible fault patterns is also significantly higher. These advantages of diagnosis under local constraints over the classical t-diagnosis approach are striking when the connectivity of the system is much smaller than the total number of processors in the system