Gossip Based Causal Order Broadcast Algorithm

摘要

Reliable group communication facility with two message ordering constraints, atomic ordering and causal ordering, is essential for distributed systems. But, as the system size rapidly increases, traditional group communication algorithms become unsuitable for very large-scale systems due to their strong reliability properties. To solve the problem, several gossip-based algorithms were presented to significantly improve scalability by ensuring the reasonably weak reliability condition. They are all designed to guarantee totally ordered delivery. However, many distributed applications such as multimedia systems and collaborative work, require causally-ordered message delivery. In this paper, we propose a Probabilistic Causal order BroadCast algorithm, PCBCast, to preserve the inherent scalability of the gossip style approach compared with the existing ones.