DEADLOCK DETECTION AND RECOVERY IN DISTRIBUTED SYSTEMS.

The deadlock detection and recovery are studied in a distributed system, where subsystems operate on equal terms without any centralized control and the number of each resource unit is one. In the system, the key is detecting the ″cyclic waiting″ and determining an interruption process, for which an algorithm is provided satisfying the following conditions. (1) Any process should terminate within a finite period; (2) when a process is determined to be interrupted, there should exist a ″cyclic waiting″ including the process; (3) the communication buffer necessary for detection and recovery amounts at most, up to that proportional to the number of processes and resources in the entire system.