Distributed software structured in terms of clients and servers is considered. Replicas of a single server are executed on separate processors of a distributed system, and protocols coordinate client interactions with these replicas. The paper describes how a system can be viewed in terms of a state machine, clients, and output devices. In this context,  Schneider  considers two representative classes of faulty behavior: Byzantine failures and fail-stop failures. The core sections of the paper present algorithms that cope with these failures. An important class of optimizations and the dynamic reconfiguration are also tackled. A separate section discusses related work.

The paper is intended for people working in the domain of distributed systems and real-time systems. It systematically presents protocols that involve replication of components using the state machine approach, although few of these protocols were obtained in this manner. The paper was received in November 1987 and the final revision was accepted in January 1990. Unfortunately, this long delay is easily perceived by the reader.