On dominating set allocation policies in real-time wide-area distributed systems

This paper investigates resource allocation policies for achieving real-time content distribution with subsecond delay bounds on the current Internet. Resource allocation in real-time systems has been concerned primarily with meeting time constraints on single processors and multiprocessors. On a single processor, the main degree of freedom available for the real-time designer is the scheduling policy. On a (partitioned) multiprocessor, a second degree of freedom is the partitioning policy. This paper explores a third degree of freedom unique to large-scale (i.e., widearea) distributed systems with non-negligible communication delays among individual nodes. We call it the dominating set allocation policy. This policy is a primary determinant of schedulability in such systems. We present some initial steps towards understanding the properties of different dominating set allocation policies in terms of resulting task schedulability. We describe an optimal dominating set allocation algorithm (subject to certain design decisions), propose a number of simple heuristics, and evaluate them using realistic Internet measurements and HTTP workload. The key contribution of this paper lies in the practical applicability of the proposed heuristics in achieving delay guarantees (with a high probability) over best-effort wide-area networks.