An efficient routing maintenance mechanism using adaptive traffic prediction

In this paper, we introduce an adaptive traffic prediction approach for optimizing the performance of dynamic routing. This approach uses the artificial neural network to predict the link traffics. The link weights can be efficiently optimized by combining the predicted available bandwidth with the current available bandwidth in order to enhance the performance of dynamic routing. We have two contributions. The first is a new adaptive prediction mechanism which is able to adapt the prediction intervals depending on the prediction accuracy in order to efficiently predict the link traffics. The second is presenting a comparative study between our prediction-based routing algorithm and two different estimation-based routing algorithms. We study three performance parameters: the rejection ratio of requests, the percentage of accepted BW and the computation time. In general, our new algorithm reduces the rejection ratio of requests, accepts more BW and has less computation time when compared to the WSP, LIOA and PFLRv.2 algorithms and based on generated and real traffic.