The timewheel group membership protocol

Abstract

We describe a group membership protocol, called the timewheel group membership protocol, for a timed asynchronous distributed system. This protocol is a part of the timewheel group communication service that supports multiple group communication semantics simultaneously. The timewheel group membership protocol is unique in several respects. First, it has been designed for a timed asynchronous distributed system model. Second, it is optimized for those failure scenarios that are more likely to occur than others. In particular, it uses a very simple and fast algorithm to recover from single failures. Furthermore, the group communication service is not interrupted, if a failure suspicion turns out to be a false alarm. Third, this protocol incurs minimal processing load during failure-free periods. In fact, this protocol does not cause any extra messages to be exchanged during failure-free periods. Finally, as a consequence of using the timed asynchronous distributed system model, this is one of the first few non-real-time membership protocols that are timed, i.e. its specification describes what outputs and state transitions occur in response to inputs and the time it takes these outputs and state transitions to occur.