Improving performance of backpressured packet networks by integrating with an end-to-end congestion control algorithm

Most of the existing congestion control schemes are classified according to where the control decision is made and what the feedback is. In backpressure mechanism, control decision is made hop-by-hop inside the network. These schemes dynamically react to congestion in a short term time scale. Hop-by-hop backpressure avoids wasting the network resources and also would reduce congestion by taking advantage of additional resources available at upstream switches and stations. Hence, it has potential for improved throughput. On the other hand, in end-to-end congestion control algorithms, congestion information is fed back, either implicitly or explicitly, and the sources decide on how to react. In this paper, we propose an integrated hop-by-hop backpressure mechanism with an end-to-end congestion control algorithm. In this combination, the end-to-end algorithm controls the network’s quasi-static behavior, while the backpressure mechanism prevents occasional congestion and packet losses that may result from traffic bursts. To integrate, we use an appropriate class of MCFC (minimum cost flow control) algorithm with desirable fairness properties. Finally, performance and fairness properties of the integrated algorithm are studied. Simulation results show that in comparison with the rate-based and window-based end-to-end congestion control algorithm, the proposed integrated algorithm is more dynamic in reacting to congestion and also more efficient in terms of avoiding loss and deadlock. Provided that stability is guaranteed, the algorithm is also quite efficient in tracking maximum throughput. The simulation results show significant performance improvement for a hybrid wired-wireless network, in comparison with TCP.