Decentralized Multi-group Communication Protocol Supporting Causal Order

Previous gossip-based protocols developed for sensor networks or various kinds of P2P networks assume that all processes in a system are interested in all events occurring while communicating among them. Though all query nodes in a large group are running the same application, they might join various multicast groups depending on their interests. However, despite the significant importance of message delivery order guarantees, little work on gossip-based protocols exists supporting multicast groups and satisfying these constraints in a scalable way. In this paper, we present an efficient and scalable causal order guaranteeing multicast protocol to use only local views supporting multi-groups. In the proposed protocol, multicast groups are defined by interests associated with the processes and join/leave messages are disseminated by gossip communication based on local views. So, group memberships are dynamically changing by their own members in multicast groups with high probabilistic reliability. Also, group ID order lists are delivered to possibly overlapping multicast groups to ensure causal order delivery. Therefore, these features of the protocol might result in its very low view management cost compared with the cost incurred by maintaining hierarchical member list for delegate groups in the previous protocols.