Use of load prediction mechanism for dynamic routing optimization

In this paper, we introduce a new efficient approach for optimizing the routing performance in MPLS based networks. The approach uses an adaptive predictor to predict future link loads. Combing the predicted link load with the current link load is an effective method in order to optimize the link weights. Our contribution is a new mechanism which uses the information of future link loads to optimize the online routing performance. A Feed Forward Neural Network (FFNN) is used to build adaptive traffic predictors which capture the actual traffic behavior. We study three performance parameters: the rejection ratio, the percentage of accepted bandwidth and the rejection ratio of re-routed requests upon link failure under two different load conditions. In general, our proposed algorithm reduces the rejection ratio of requests, achieves higher throughput and reroutes more requests upon link failure when compared to the DORA and LIOA algorithms.