Prediction-based Decentralized Routing Algorithm

Abstract

We introduce a new efficient routing algorithm called Prediction-based Decentralized Routing algorithm (PDR), which is based on the Ant Colony Optimization (ACO) meta-heuristics. In our approach, an ant uses a combination of the link state information and the predicted link load instead of the ant's trip time to determine the amount of pheromone to deposit. A Feed Forward Neural Network (FFNN) is used to build adaptive traffic predictors which capture the actual traffic behaviour. We study two performance parameters: the rejection ratio and the percentage of accepted bandwidth under two different network load conditions. We show that our algorithm reduces the rejection ratio of requests and achieves a higher throughput when compared to Shortest Path First and Widest Shortest Path algorithms.