We introduce the timed asynchronous distributed system model to describe existing asynchronous distributed systems subject to unbounded processing and communication delays, failures and recoveries. We then describe five increasingly strong specifications for processor-group membership services in timed asynchronous systems subject to partitioning. We also propose five distributed protocols that implement these specifications despite arbitrary numbers of crash/performance processor failures and omission /performance communication failures, and prove their correctness. Finally, we show how two of the protocols can be adapted to implement a highly available processor leadership service that ensures the existence of at most one leader at any point in real-time.

Abstract. We derive the speci cation of a primary partition and a partitionable fail-aware node membership service in a top-down fashion. The derived speci cations are fail-aware in the sense that each client of a membership server can learn if the server currently provides its standard semantics or an exception semantics because too many failures have occurred. We rst propose the speci cation of an ideal membership service and then transform this ideal speci cation step by step to derive the two fail-aware speci cations that are implementable in timed asynchronous systems. In each step we address an implementation problem orachange in the system/failure model. 1