Termination detection using causality and vector time

A problem unique to distributed systems is the need to determine that a computation is terminated. We use vector clocks as a basis for a termination detection protocol in which each process is only contacted once following termination. In addition we also present a definition of termination that does not depend on global state. This definition is based on the causal relationships between events and is stated in terms of Lamport's "happens before" relation. Because the protocol is also specified in causal terms, this definition can be easily used to formally argue the correctness of the algorithm.