Research On Network Load Balancing Strategy Of SDN-based Cloud Data Center

In recent years,the application of cloud computing and big data analytics has developed rapidly,and cloud data centers are important infrastructures to support them.In cloud data centers,data related to cloud computing and big data analytics is mainly transferred between servers and storage systems within the data center,consuming a large amount of network bandwidth resources and posing a huge challenge to the network topology and bandwidth resource allocation of the data center.This thesis aims to optimize the allocation and utilization of bandwidth resources in cloud data centers and improve network service performance by implementing network traffic load balancing strategies through the analysis of network traffic characteristics and network resource consumption within cloud data centers,taking advantage of global state awareness and dynamic policy deployment in SDN networks.The main works of the thesis are as follows.(1)The network traffic characteristics of SDN-based cloud data centers were studied.Through the analysis of the network topologies and traffic characteristics of cloud data centers,the characteristics of the traffic composition and transmission range of cloud data centers are summarized;the technical principles and role of SDN technology in the virtualization of resources in cloud data centers,the dynamic and elastic provisioning of network resources,the agile and flexible deployment of network traffic control strategies are discussed.Then,the general idea and approach of SDN-based network traffic control and management in cloud data centers are summarized.(2)A software-defined load balancing strategy SDLB for cloud data centers is proposed.In a typical Fat-Tree topology as a research scenario,the traffic characteristics within the cloud data center network are analyzed.The elephant flows that consume the main bandwidth resources are screened.Synthesizing the two factors of link distance and link bandwidth utilization,the SDN controller is used as the policy control center,sensing the global link state,determining the load balancing strategy.The strategy is deployed to the corresponding switches to realize the rerouting of the elephant flows.And the overall traffic load balancing is achieved.(3)A PSO-based elephant flow re-routing algorithm EFRA is proposed,which takes advantage of the simple and fast convergence of PSO to cope with real-time route adjustment during load balancing,with two constraints of path distance and link bandwidth utilization introduced in the elephant flow re-routing calculation.Using random inertial weighting factor,the ability of PSO algorithm to solve global optimal solution is improved.So the proper elephant flow re-routing solution is obtained.(4)Experimental scenarios of data center with typical Fat-Tree topology is constructed through the Mininet platform.Experimental validation of the SDLB strategy as well as the EFRA algorithm is done with the comparison with the equivalent multipath routing algorithm ECMP and the Global First Fit routing algorithm GFF.The results show that the SDLB strategy and EFRA algorithm proposed in the t can effectively improve the load balancing performance of cloud data centers,and the network performance metrics such as network throughput and average bisection bandwidth are improved.