A fault-tolerant distributed algorithm for termination detection using roughly synchronized clocks

A fully symmetric and distributed solution to the termination detection problem is presented along with a proof of its correctness. Each of the processors in an asynchronous (or synchronous) network has a physical clock which is roughly synchronized with all other processors' physical clocks. The control messages that are timestamped using roughly synchronized clocks are circulated around an unidirectional (logical or physical) ring of all processors to detect termination of the underlying computation. The algorithm detects termination along with the correctness of the result of the underlying computation even in the situations when the processors fail and recover. The algorithm also works correctly even after the failure of clock synchronization. This algorithm involves considerably less number of messages to detect global termination in any network (synchronous or asynchronous) compared to other similar algorithms in the literature.