Enhanced TCP Vegas for asymmetric networks

It has been shown that Vegas underutilizes the available bandwidth in asymmetric networks in which the bottleneck is in the backward path rather than in the forward path. In this paper we propose a solution by making use of the ecn/spl I.bar/to/spl I.bar/echo/spl I.bar/ bit. Based on it, the congestion probability in the backward path is estimated. We show that in case of network congestion, the change of RTT is proportional to the change of congestion probability. Hence, when backward congestion occurs, by monitoring the change of congestion probability we can estimate the change of RTT. Then the actual RTT (before congestion) can be estimated and we use it to calculate the actual throughput. We also show that the proposed method is applicable when congestion occurs both in forward and backward path. Simulation result gives some illustrative examples.