HiScamp: self-organizing hierarchical membership protocol

Gossip-based or epidemic algorithms rely on a peer-to-peer model for dissemination of multicast messages, and are simple, scalable and reliable. However, traditional gossip-based protocols suffer from two major drawbacks: (i) they rely on each peer having knowledge of the global membership and (ii) they are oblivious to the underlying network and impose a high load on core router links. In this paper we present a self-organizing hierarchical membership protocol which attemps to solve these two issues. Nodes organize themselves into clusters reflecting the topology, and obtain partial views of the membership both within and outside the cluster. The size of the partial views is tuned automatically to achieve high reliability. Gossip messages are targeted mainly within clusters, thereby reducing network load.