Congestion Control by Heuristics in High-Speed Networks Using ANN

Abstract

A primary reason for Congestion in communication networks is the presence of preprocessing and postprocessing queues (buffers) in routers and switches. Congestion in the networks reduces the quality of service (QoS), hence causing delays and loss of data. Most of the existing congestion control techniques are incapable of controlling the congestion that might occur in the future, as they do not use any prediction algorithms. To overcome this inability, in this paper, a new heuristic congestion control technique is proposed to determine the present status of congestion in the network and also predict the congestion in future. The proposed heuristic congestion control technique maintains a historical database, containing the congestion control parameters such as round-trip time (RTT), average queue size (AQS), available bandwidth, and network speed. Based on this historical database, the current status of congestion is determined and also its future value is predicted in networks with the help of an AI technique using a feed-forward backpropagation neural network (FFBNN). Finally, the sending rate of the packets to the network is adjusted using these congestion control parameter values, so as to reduce congestion in the network. An optimal congestion window is designed using an optimization algorithm called evolutionary programming (EP) algorithm. Simulation results have shown that the proposed heuristic congestion control technique efficiently controls the congestion in the networks, increases the throughput by 20 %, reduces the propagation delay by 22 % when compared to the existing technique [10], and also improves the performance of the network by 20 %.