High availability multiprocessor systems

System availability is of primary interest to a wide variety of compute-intensive application areas, such as scientific and financial modelling, air traffic management, medical computing, and process control. Traditionally, this has been achieved by replicative fault-tolerance, but this is grossly expensive. The advent of the transputer has provided the opportunity for use of techniques employed in local area networks to migrate down into more tightly-coupled computing systems. The paper describes the development and transputer implementation of an alternative to replicative fault-tolerance which offers not only high-availability, high-performance computing, but also features lower cost, graceful degradation, and the potential for use in safety-critical environments.<>