Differentiating link state advertizements to optimize control overhead in overlay networks

Routing in overlay networks typically involves engineering an overlay topology on top of the Internet to balance traffic along overlay paths so that quality and/or resilience of delivered services are improved. It can be used to reduce latency for delay-sensitive applications. It then consists in selecting, for any pair of nodes, an intermediate overlay node which reduces the latency on this one-hop overlay path against the latency on the direct overlay path between them. In this paper, we propose to optimize the overhead generated by the overlay route computation mechanism by introducing a differentiation between the nodes that are highly used as relay and those that are not. Our approach relies on disseminating at a high frequency the link states with the identified sub-set of nodes and at a lower frequency all the link states. We conduct large experimentations on PlanetLab to evaluate the trade-off between the performances in terms of RTT gain and the reduction of the control overhead compared to the state of the art.