Cooperative resource discovery and sharing in group communications

In this dissertation we study four areas where members’ cooperative behavior in discovering and sharing network resources can be beneficial to achieving various objectives in group communication. We propose a framework for discovering the topology of a shared multicast tree based on a novel fan-out decrement mechanism analogous to time-to-live (TTL) decrementing in IP. The proposed algorithm for topology discovery is based on the matrix of path/fan-out distances among session members. We exhibit sufficient conditions for topology discovery based on a reduced distance matrix, and propose a practical protocol to acquire this information. Additionally, we show how the same approach permits nodes to discover the multicast distribution tree associated with members within their fan-out/TTL scoped neighborhoods. This permits one to reduce the computational costs while making the communication costs proportional to the size of neighborhoods. We then present a novel distributed and scalable framework to support on-demand filtering and tracing services to defeat distributed Denial of Service attacks. Our filtering mechanism is designed to quickly identify a set of boundary filter locations so that attack packets might be dropped as close as possible to their origin(s). We argue that precisely identifying the origins of an attack is infeasible when there is only a partial deployment of tracing nodes - as is likely to be the case in practice. Thus we present a tracing mechanism which can identify sets of candidate nodes containing attack origins. Both mechanisms use multicasting services to achieve scalable, responsive and robust operation. Next we propose a topology-sensitive subgroup communication (TSC) mechanism to support efficient subgroup communications in large-scale multicast applications. Our TSC mechanism exploits spatial locality among members communications within a given subgroup, and enables members to autonomously build a TSC forwarding structure consisting of multiple unicast and scoped multicast connections. This can completely eliminate the need to create additional multicast sessions while minimizing the exposure of receivers to unnecessary packets. Simulations of this approach suggest that TSC mechanisms perform well for diverse densities and distributions for a subgroup’s nodes. Finally we propose a method to quickly distribute large files across distributed nodes. Our Adaptive FastReplica (AFR) mechanism exploits path diversity among the origin and receivers and adaptively balances loads across overlay paths. Since our approach uses a fixed overlay structure but then adapts the loads across paths, the control overhead associated with constructing and maintaining overlay structure, typical of application-level multicasting solution, is reduced. Based on our experiments with a prototype implementation over the Internet, we demonstrate its efficiency at minimizing the overall replication time.