A Membership Protocol for Multi-Computer Clusters

Distributed applications need membership services to know which of their components are active or have failed. The Cluster Membership Monitor (CMM) provides membership services at the machine level. As its name suggests, the target environment for this membership service is a cluster of computing nodes, where the CMM checks the current state of all machines that have a CMM instance running on them. All monitors run a distributed membership protocol with an integrated timer- based failure detector. The aim of these monitors is to notify all membership changes to a set of software packages that previously requested their services. As a result, all these packages know about the current configuration of th e cluster where they are running and may adapt its state to this configuration. The protocol has been used as the basis for the development of a reliable communication layer and as an assistance service for the reconfiguration of an object request brokerwhich provides the remote invocation support for the object-oriented applications running in the cluste r.