Research On Load Balancing And Elephant Flow Dynamic Rerouting Based On SDN Hierarchical Controllers

With the ever-expanding scale of network and the emergence of cloud services,traditional network infrastructure has become more and more complex,and operators and enterprises have found that their network infrastructure has been difficult to meet new service requirements.Software Defined Network(SDN)provides operators and enterprises with new mechanisms and tools that enable them to easily deploy services while reducing capital expenditures and operational costs.Therefore,SDN has attracted global attention.However,the centralized control mode of the SDN controller makes the load on the control plane too large.The large number of stream requests of the switch and the message exchange between the switch and the controller can easily cause the overload or load imbalance of the controller.Data center networks can provide high-performance communications for a variety of applications,which also allows more and more computing and storage resources to be transferred from traditional Internet to the data center networks.However the deployment of large-scale bandwidth-intensive services has led to a rapid increase in data center network traffic,which has brought severe challenges to the data center network.How to improve the utilization of network resource,reduce operating costs,and effectively control the data center network traffic has become an urgent problem to be solved.The centralized control and programmable characteristics of SDN play an important role in the collection and management of traffic,so SDN is increasingly being used in data center networks.However,data center networks still face many challenges in terms of traffic conflicts and link congestion.The main research contents of this thesis are as follows:(1)Aiming at the problem of high communication overhead between controllers and low controller throughput in SDN multi-controller load balancing process,an SDN-based hierarchical controller load balancing mechanism LBMHC(Load Balancing Mechanism for Hierarchical Controllers based on SDN)is proposed.LBMHC adopts a layered multi-controller architecture to achieve load balancing through cooperation between the super controller and the domain controller.When the domain controller sends the load information to the super controller,a suppression algorithm for load information transmission is proposed.The algorithm uses the load threshold segmentation and defines information collection intervals to avoid periodic transmission of information between the domain controller and the super controller.At the same time,the LBMHC sets the corresponding priority scheduling mechanism according to the load severity of the domain controller,preferentially processes the most overloaded domain controller,and selects multiple matching migration criteria according to the priority from the switch controlled by the overload domain controller.Then,the selected switches are respectively migrated to the corresponding higher priority target domain controllers,effectively avoiding the situation where multiple switches connected to the overload domain controller are simultaneously migrated to the same target domain controller.Finally,the experimental scene is simulated by Mininet.The simulation results show that the proposed mechanism can effectively reduce communication overhead,improve system throughput and have a better load balancing effect.(2)Aiming at the problem of network congestion caused by elephant flow hashing to the same path in data center network and conflict between elephant flow and mouse flow in the same path,a dynamic rerouting scheme for elephant flow based on SDN hierarchical controller is proposed.The scheme is based on the hierarchical multi-controller architecture proposed in LBMHC.Firstly,a collaborative elephant flow detection method is proposed,which combines sFlow detection with controller statistical flow detection to identify elephant flow.Then,the Yen algorithm is used to obtain K short paths based on the number of hops,and the cost function model is used to calculate the lowest cost short path according to the link bandwidth utilization and the network delay.After that,the traffic balance is used to determine whether the traffic distribution is balanced.If it is unbalanced,the link bandwidth utilization rate and the bandwidth ratio of elephant flow are used to dynamically reroute the elephant flow that may cause link congestion.Finally,the Fat-Tree topology is simulated by Mininet,the simulation experiments show that the proposed cooperative elephant flow detection method is accurate and high in detection efficiency,and the scheme of dynamic reroute for elephant flow in data center can effectively improves the link utilization rate,reduce transmission delay and achieve better network performance.