Artificial Intelligence Empowered Load Balance Routing Strategy For Software Defined Network

With the emergence of traffic services,such as the mobile networks and the Internet of Things(IoT),more and more devices are connected to the Internet for data transmission,which will bring about an explosive growth of traffics.Meanwhile,the network will be able to share physical infrastructure to provide information services,such as news releases,online broadcasts,remote medical care,real time communication,etc.Hence,traffic services require high Quality of Service(QoS),including delay,jitter,bandwidth,Packet Loss Rate(PLR),etc.Traditional routing algorithms always adopt "best effort" strategies,which cannot meet the diversified QoS constraints brought by various services.The data-driven intelligent routing algorithm is difficult to implement because of its poor interpretability,compatibility and robustness.Given the source and destination of a traffic request,this research hopes to find a load balance routing strategy,where machine learning can gift its intelligence and traditional routing decision can ensure its reliability.This paper proposes two algorithms for network load balance:Machine Learning Aided Load Balance Routing Scheme Considering Queue Utilization(MLQU)and QoS-oriented Adaptive Routing Scheme Based On Deep Reinforcement Learning(QAR).The main contributions of this paper are listed as follows:(1)This paper identifies the relationships between the QoS parameters and different resources relying on a jitter graph-based network model and a Poisson process-based dynamic traffic model.Meanwhile,Principal Component Analysis(PCA)method is introduced into the extraction of topology,which can effectively represent the topology connection of each router.(2)The MLQU algorithm perceives the future network situation with traffic prediction module and determines an efficient,flexible,as well as intelligent routing scheme.Moreover,the neural networks explore the relationship between traffic requests,network topology and routers'queue status to predict the next queue utilization.(3)This paper proposes a QoS Routing Strategy with Resource Allocation(QRRA)based on the queueing theory by a low computational complexity greedy algorithm.(4)The QAR algorithm uses the Deep Deterministic Policy Gradient(DDPG)algorithm for load balance routing,which can better solve the problem of real-time dynamic changes in network status and data traffics.Finally,simulation results show that our proposed algorithms outperform the traditional strategies in terms of packet loss rate,throughput and queueing delay.