End-User Policies for Predicting Congestion Patterns in Data Networks
Laurent Massoulié, Peter B. Key, Koenraad Laevens · 2000
In this paper we propose a scheme for detecting and predicting congestion patterns from network feedback. The focus here is on binary feedback messages, such as loss events or ECN marks. Simulations show that, when a fraction of end users modify their congestion control based on these estimations, oscillations in network buffer occupancy otherwise induced by TCP users are significantly reduced. These oscillation reductions benefit both TCP users and the users applying the proposed scheme, since both types of users receive a smaller proportion of ECN marks. 1.