Research On Network Traffic Scheduling Mechanism Of Edge Data Centers

The explosive growth of network traffic has put forward higher requirements for the development of data centers.Compared with traditional centralized data centers,edge data centers bring users a high-quality network experience due to their proximity to users.However,the current edge data centers still have the problems of unreasonable traffic scheduling and low resource utilization.Therefore,this dissertation carries out research on network traffic scheduling in edge data centers aiming at solving the problems existing in edge data centers.The main research of this dissertation is as follows:(1)First,we use software-defined network technology to build a system architecture that can obtain network status information in real time,providing basic support for subsequent routing optimization.Secondly,we use the perception and decision-making capabilities of deep reinforcement learning to explore the potential connections in the edge data center network,and achieve the purpose of scheduling the network traffic of the edge data centers by considering the bandwidth and load balancing degree of the edge data centers.A routing optimization algorithm(TRO)based on twin delayed deep deterministic policy gradient is proposed.The algorithm outputs a weight matrix,which provides a basis for traffic routing and achieves the purpose of network traffic scheduling in edge data centers.Finally,the experimental results show that TRO effectively improves the bandwidth utilization of edge data centers,reduces network delay,and improves the degree of load balancing.(2)In order to better optimize the network traffic scheduling strategy of the edge data center,we conduct research on the network traffic prediction of the edge data center.We apply the attention mechanism to the Long Short-Term Memory(LSTM)and proposes a traffic prediction method based on attention and LSTM(A-LSTM).Several comparative experiments are conducted through public datasets and datasets collected from edge data centers.The experimental results prove that A-LSTM can accurately predict the traffic of edge data centers.On this basis,an edge data center routing optimization algorithm(PTRO)based on A-LSTM and TRO is proposed.The algorithm uses the prediction results of the access layer switches in the edge data center,and performs secondary processing on the weight matrix output by the TRO algorithm,so as to improve the network bandwidth resource utilization of the edge data center.The experimental results show that PTRO has a faster convergence speed and further improves the performance of edge data centers.