Shmuel Safra’s Termination Detection Algorithm

Abstract

Along with the rise of large-scale distributed systems, a number of new-problems and algorithmic paradigms saw the light. One of these was the problem of so-called quiescence detection, which led to the wider class of so-called distributed snapshot algorithms. Roughly speaking, distributed snapshot algorithms have to detect, in a distributed fashion, whether a system has entered a state in which no relevant things will happen any more. What is relevant very much depends on the specifics of the problem. For instance, in a self-stabilizing system, i.e. a system that stabilizes towards some pre-specified state, one might be interested in whether that state has been reached. The so-called termination detection algorithms, which were the first algorithms of this class to attract the attention of computing scientists, are to detect whether a distributed computation proper has arrived at a final state, i.e. has terminated.