Multi-objective optimization of two control strategies for congestion avoiding in computer network

Abstract

Network congestion is a phenomenon strongly impacting the real level of efficiency expected from the modern network environment. It has direct impact on reliability increase and high, stable throughput. It's also one of the main reasons of the end to end delay increase. These characteristics define the quality of communication channels. Optimizing network nodes configuration for only one of the given features can exacerbate the other parameters. This paper focuses on avoiding and alleviating network congestions using multiobjective optimization. This optimization process is used to adjust the controller gain and optimal reference signal. Unlike in other presented approaches, in this case, a dynamical, discrete, non-stationary model of communication channel is applied. It reflects delay conditions in a real environment, which are varying in time. The advantage of such approach rises from the comparison with analogue stationary models, what is also discussed in this research. Two different control strategies have been chosen to be subjected to the optimization process. The first one uses the constant, optimized reference value. The second one is using an adaptive reference value. The proposed strategy of congestion control is adjusting parameters of the presented models to alleviate undesirable results of sudden, network condition changes. It is obtained by maximization of available bandwidth usage combined with the minimization of buffer utilization. This approach supports avoiding harmful congestion effects like retransmission, packet dropping and high network delay, which eventually cause network throughput degradation.