New Schemes for Predictive Congestion Control

摘要

The Predictive Congestion Control framework, as proposed by Ko, Mishra, andTripathi, applies to high-speed, wide-area communication networks. The central assumption of the framework is that the link propagation delay, a deterministic quantity depending on fiber length, dominates all other types of delay in the networks. Within the framework, the authors developed four schemes, the static, optimistic, pessimistic, and heuristic, which are summarized here. This thesis presents two new schemes and compares their performances to those of the original four schemes. One new scheme, the square-root queue scheme, adapted from a dynamic window scheme of Mitra and Seery, attempts to equate the square of the average buffer occupancy and the product of the output rate and the link propagation delay. The second new scheme, the equal-risk principle scheme, motivated by keeping the probability of packet loss small, compares available space to the requested output rate. The schemes were tested on two cases in simulations: one virtual circuit with cross-traffic sharing one link, and two virtual circuits having one link in common. The measures used to judge performance are packet-loss ratios for the congested link, for the internal network, and for the end-to-end virtual circuit. Based on these measures, the square-root queue scheme and the equal-risk scheme perform as well as but not significantly better than the static scheme.