LARK: a light-weight, resilient application-level multicast protocol

Application-level multicasting (ALM) has attracted a significant amount of attention, as it is a convincing alternative over traditional IP multicasting. We present a simple, light-weight, yet scalable, ALM protocol, called. LARK, that allows the formation and maintenance of overlay topologies in a completely distributed fashion while maintaining only O(1) state at each node and ensuring robustness in the presence of a large number of node failures. Conceptually, members self-organize into cliques, where a clique is a cluster of end-hosts in which each end-host is aware of, and exchanges state with, every other end-host in the cluster. No control message is exchanged for clique maintenance beyond the necessary state update among members belonging to the same clique. In addition, members are allowed to peer with randomly selected members belonging to other distinct cliques. This ensures that in the event that one or more members leave or fail, the other members can re-join the group at other peers they are aware of. We elaborate on the components of LARK and derive certain theoretical bounds on its performance. We also validate our design through simulations.