A dynamic bandwidth allocation algorithm based on neural network prediction-correction model and software defined TDM-PON

Abstract

In the future, with multiple services and large-capacity access network scenarios, the network load is often high but the bandwidth is limited. On the situation, based on the software-defined TDM-PON access network architecture and network traffic prediction-correction model, a dynamic bandwidth allocation algorithm is proposed. In the algorithm, a prediction model is used to predict traffic information and a correction mechanism is used to correct the prediction model. After analyzing the global information of the network, the algorithm provide corresponding bandwidth management policies based on business priorities according to different network load conditions. We compare this algorithm with IPACT algorithm, unused prediction algorithm and neural network prediction without correction. It proves that the algorithm guarantees the service quality requirements of different priority services when the bandwidth is limited and the network load is high, and it performs better in terms of average packet delay, bandwidth utilization, etc. Simulation shows, compared with the traditional strategy, the average packet delay is reduced by 70%, and the bandwidth utilization is increased by 19%.