A Customized Communication Subsystem for FT-Linda

Distributed fault-tolerant systems usually impose much stronger requirements on the underlying communication protocols than do applications developed without fault-tolerance in mind. That is true, for example, of applications composed of processes replicated on multiple hosts, where all replicas must keep the same view of the state of the communication. This paper describes how the communication substrate for a speci c application with strong communication requirements was developed. The application, the runtime system for a fault-tolerant version of the Linda language called FT-Linda, requires a communication substrate capable of providing ordered atomic multicast, failure detection and membership services. The implementation relies on a new framework for the composition of event-driven microprotocols that is used with the x-kernel.