Energy-Efficient Routing And Traffic Scheduling Mechanism In Software Defined-DCN

To support the continuous development of cloud computing,Internet of Things and other technologies,more and more data centers have been established.However,the equipment utilization rate of data center network is low,and there are often many redundant links and excessive link bandwidth,which lead to low energy efficiency.SDN technology has the characteristics of global view and unified control.Operators can use the characteristics to design and implement traffic engineering solutions in a more fine-grained way to better manage network resources.Then energy efficiency can be effectively improved.Therefore,it is of great significance to study the energy efficient routing and flow scheduling of SDN-based data center network.The resource allocation mechanism of SDN-based data center network faces the following challenges:Firstly,the network needs to give priority to the energy consumption at the device level,and the energy consumption at the link rate level can not be ignored.The current research based on adaptive link rate technology mostly adoptted the transmission mechanism with constant rate,while the transmission mechanism with variable flow rate can fully improve the energy efficiency of link rate level from the two dimensions of time and space.Secondly,while reducing network energy consumption,operators need to effectively improve users' satisfaction with the network.The network carries a variety of services,which can be roughly divided into bandwidth sensitive and delay sensitive ones;at the same time,the network traffic has unique characteristics,such as the distribution characteristics of elephant flow and mouse flow.Therefore,when designing energy-saving routing and scheduling mechanism,we need to refer to the service performance requirements and traffic characteristics to reduce energy consumption and effectively improve user satisfaction.Thirdly,SDN network includes data plane and control plane.On the one hand,the controller load balancing problem is introduced,and in band control mode,the data path and the control path share physical resources,and the energy-saving routs for data flows and control information need to be calculated to optimize the topology;on the other hand,flexible flow scheduling mechanism can effectively improve the performance of data plane,such as the average completion time of the flows.In view of the above problems,the main research contents of this paper are as follows:(1)Aiming at the problem of energy-saving topology optimization for control plane performance optimization,a cooperative energy-saving routing algorithm based on DRL(deep reinforcement learning)is proposed.Firstly,for the invband control mode,the energy consumption and the load balance of the control plane are taken as the joint optimization objectives.Secondly,The DRL algorithm is designed,and the data path and control path are obtained according to the experience training,then the energy-saving topology is obtained,and then the idle devices are dormant.Finally,the simulation results show that the proposed algorithm can effectively improve the load balance of control plane on the basis of energy saving.(2)Aiming at the problem of low energy efficiency in the scenario of energy-saving topology determination in data center network,an energy-efficient routing algorithm based on dynamic programming is proposed.Firstly,based on the energy consumption model of link rate level,for the flows with the deadline requirements,which is the key factor affecting the performance of data exchange,and the deadline requirement is taken as the QoS limiting condition,and the energy-saving QoS routing optimization problem model is established.Secondly,based on link rate adaptive technology,a transmission mode with variable flow rate and bandwidth sharing are proposed.The flows are sorted according to the order of the deadline,and the Floyd Warshall dynamic programming method is improved to calculate the optimal energy-saving routing scheme for each flow.Especially in the case of route selection failure,on the premise of minimizing the overhead,the controller selects part of the flow for rerouting to improve the network flow acceptance rate and throughput.Finally,the experimental results show that the routing algorithm can effectively reduce the network energy consumption,and the local rescheduling algorithm can improve the network flow acceptance rate without significant increase in energy consumption.(3)Aiming at the energy-saving flow scheduling problem in the scenario of network energy-saving topology determination and routing fixed,a dynamic energy-saving flow scheduling mechanism based on flow feature classification is proposed.Firstly,for the flows with deadline requirements,the energy-saving flow scheduling problem model is established based on the link rate level energy consumption model.Secondly,the flow classification scheduling scheme based on the features of elephant flow and mouse flow is proposed,in which the dynamic scheduling mechanism is only adopted for elephant flows.At each time,there is no need to predict the information of the future elephant flows.First,the current and previously unfinished elephant flows are merged,and then the Most-Critical-First algorithm is used for unified scheduling.Finally,simulation results show that the proposed algorithm has stable and effective energy saving effect and achieves the balance between energy saving and performance.(4)Aiming at the energy-saving flow scheduling problem for data plane performance optimization,a DRL energy-saving flow scheduling mechanism is proposed.Firstly,for the flows with deadline requirement,the minimum network energy consumption and average completion time of network flow are taken as the joint optimization objectives.Secondly,by making full use of the advantages of deep deterministic strategy gradient algorithm in solving dynamic and continuous control problems,the flow scheduling mechanism with variable flow rate is designed.Finally,experimental results show that the proposed algorithm can effectively reduce the average completion time of the flow while reducing the energy consumption of the link rate level.