This thesis is concerned with the distributed fault-diagnosis and its theoretical properties for fault-tolerance in distributed network systems. It is assumed that a DNS(distributed network system) has fully distributed control ; each node in DNS has its own operating system and memory. a DNS is defined to be t-fault self-diagnosable if every nonfaulty node can correctly identify all faulty and nonfaulty nodes, and faulty edges between pairs of all nonfaulty neighbors in the system DNS, provided that the total number of node failures and edge failures does not exceed it.
We present the necessary and sufficient conditions for t-fault self-diagnosable systems and construct optimal t-fault self-diagnosable systems. A diagnosis algorithm for the t-fault self-diagnosable system is given.