Reducing Packet-Loss by Taking Long-Range Dependences into Account

摘要

We show that the "fractal" behavior of Internet traffic can be efficiently and practically employed to significantly reduce packet-loss. Thanks to recent advances in the theory of self-similar processes, we define the probabilistic congestion of a link, based on an estimated computation of the packet-loss probability over that link. This congestion parameter allows valid predictions on the future behavior of the network, on which one can base efficient routing strategies. We show how to implement the computation of the probabilistic congestion, and we illustrate several applications for improving unicast and multicast protocols.