Agreement on the group membership in synchronous distributed systems

Abstract

When a group of processors in a distributed system cooperate with each other on processing of a common task, it is often necessary for the non-faulty processors to have a mutually consistent knowledge of the set of processors that can be considered to be non-faulty. The set of non-faulty processors in the group — known as the group membership — will change for example when a processor crashes or when a crashed processor, after restart, joins the group. These changes should be known by all non-faulty processors as quickly as possible within a known bounded time interval. We present an algorithm by which non-faulty processors of a group of bounded size will be able to maintain a consistent and timely knowledge of the group membership. Processors in the group are assumed to execute the algorithm in a synchronous manner and at periodic intervals or cycles of some fixed length. In an execution of the proposed algorithm, every non-faulty processor knows of any processor failure within at most two cycles following the cycle in which the failure occurred, and a restarted processor can join the group in two cycles. At most less than half the number of processors are assumed to fail in any three consecutive cycles.