Research On Energy-saving Routing Optimization Of Software-defined Data Center Network Based On Traffic Awareness

As an important infrastructure supporting big data and cloud computing in the network,data centers have always been concerned about their energy consumption.As the energy saving of data centers requires more flexible management,the emerging technology software-defined network that allows flexible control of network equipment brings new opportunities for energy-saving routing optimization of data center networks.Software-defined network(SDN)is a new network architecture in which the control plane and the data plane are separated as the key design principle.It has the characteristics of flexible and logically centralized control;it takes advantage of the centralized control of the SDN architecture to combine big data analysis and network functions.Programming features to build energy-saving networks to reduce energy consumption in data centers is of great significance to the realization of a green economy,environmental protection and sustainability.There are two main problems in the existing energy-saving routing algorithms based on traffic aware: one is that the existing traffic aware methods cannot provide high-precision traffic prediction,resulting in poor subsequent energy-saving routing optimization effects;second,the existing energy-saving routing The algorithm optimization goal is relatively simple,and its applicability to complex networks is relatively poor.In response to the above problems,this article has conducted an in-depth study on energy-saving routing issues in data centers,using the advantages of SDN centralized control,and on the basis of existing research work,proposed an intelligent-driven software-defined data center network energy-saving routing architecture.Using network data,the graph convolutional network is used to accurately predict the traffic in the data center,and the deep reinforcement learning model is used to achieve real-time dynamic energy-saving routing optimization.First of all,in view of the problem that most current researches only focus on static optimization of software-defined data center energy-saving routing,this paper proposes an intelligent-driven data center energy-saving routing architecture that supports the integration of "traffic perception,routing decision-making,and dynamic adaptation" design.Secondly,in response to the problem of low traffic perception prediction accuracy,this paper proposes a graph convolutional gated neural network(Graph Convolutional Gated Recurrent Network,GCGRU),adding a graph convolution layer can fully perform the spatiotemporal characteristics of SDN data flow Learning improves the model's perception accuracy of network traffic,and is used for subsequent dynamic prediction of link utilization thresholds and energy-saving routing optimization.Third,for energy saving,load balancing,delay and other performance indicators that are difficult to balance,this paper is based on the double-threshold method of link utilization to ensure that not only energy saving is considered when scheduling flows,but load balancing is also taken into account;use in-depth enhancement Learn the advantages of being good at solving decision-making problems,make full use of the competitive deep Q network based on priority experience playback,treat the routing task as a discrete control problem,and design a reward function that weights network performance parameters.The design of the reward function integrates important performance parameters such as energy saving,load balancing,and delay,and fully considers the performance of the network while considering the energy saving effect,and realizes dynamic adaptive energy-saving routing optimization.Finally,use Mininet and POX to build the experimental simulation environment of this article.First,by comparing with the typical algorithm on the measured GEANT data set,first verify the predictive ability of the TSACN model.The experimental results show that the model is in the mean absolute percentage error(MAPE)Compared with the GRU and GCN models,the above is reduced by 0.52% and 1.54%,respectively.Then compare the TA-PDDQR(Traffic-Aware Prioritezed Dueling Deep Q Network Energy Routing)algorithm proposed in this article with a typical energy-saving routing algorithm.The results show that the algorithm can save up to 12 Watt on average switch energy consumption and 40% of links.The above is excellent in terms of average response delay,improvement of network resource utilization and throughput,which proves that the framework of this paper can better achieve the balance of energy saving and network performance based on GCGRU prediction.