The Implementation of Reliable Distributed Multiprocess Systems

A method is described lor implementing any system by a network of processes so it continues to function properly despite the failure or malfunction of individual processes and communication arcs; where "malfunction" means doing something incorrectly, and "failure" means doing nothing. The system is defined in terms of a sequential "user ma chine", and a precise correctness condition for the Imple mentation of this user machine is given. An algorithm lo im plement the user machine in the absence of malfunctioning, and a rigorous proof of its correctness, are given for a net work of three processes with perfect clocks. The generaliza tion to an arbitrary network of processes with imperfect clocks is described. It is briefly indicaled how malfunctioning can be handled by adding redundant checking to the imple mentation and Including error detection and correction mechanisms in the user machine. Associates, and is currently at SRI International. He has done reserach in analytic partial differential equa tions, and in various aspects of con current processing. His current inter ests include the theory of concurrent processes, the foundations of tiuan-tum mechanics, and applied oriental philosophy.