Research On Load Balancing Algorithms Of Data Center Network Based On SDN

In recent years,the rapid rise of cloud computing and other emerging technologies has brought huge network traffic pressure to the data center.It is difficult for traditional network architecture to efficiently schedul e traffic,so that it cannot meet the performance requirements of its applications.Software Defined Network(SDN),as an emerging Network architecture,has completely overturned the traditional Network traffic scheduling method.It separates the control a nd forwarding of network data so that it has the characteristic of centralized architecture and can grasp the global network view.This provides a good idea for data center traffic scheduling and load balancing.This paper focuses on the load balancing of data center network in SDN environment.According to the classification criteria of load balancing algorithm,this paper separately proposes two improved algorithms for the problems that the deterministic algorithm has uneven scheduling of large flows,whi ch easily causes network congestion,and the problems of non-deterministic algorithms such as slow convergence and premature search stagnation.1.Large-flows Distribution Load Balancing Algorithm.This algorithm takes the mean square error of the load of the whole network link as the optimization objective.Firstly,the Yen algorithm is used to calculate the initial candidate path set of the flow,and then the path with the highest available load is selected to narrow down the corresponding candidate path range.Finally,the path with the highest available load is taken as the final flow scheduling scheme.The simulation results show that this algorithm has better performance than other classical deterministic algorithms.2.Ant Colony Improvement Load Balancing Algorithm.The algorithm takes the average link bandwidth utilization as the optimization objective.Firstly,according to the load of network links,the pheromone of the initial whole network links is differentiated.Then,using the idea of weighted evaluation,a synthetic heuristic function is designed.Secondly,an optimal and worst global pheromone updating rule is proposed based on the link pheromone updating rule and the reward and punishment system.Meanwhile,a phased pheromone volatility coefficient is designed according to the segmented strategy.Finally,in order to prevent the pheromone between links from becoming too different with the increase of the number of algorithm iterations,the pheromone value on the link is limited to a certain range.Compared with other deterministic and nondeterministic algorithms,this algorithm has certain advantages in performance indexes.In addition,this paper designs and completes the whole load balancing system of data center network based on the above algorithms,which mainly includes topology discovery module,information acquisition module,flow detection module,load balancing monitoring module,path decision module and flow table installation module.Among them,the path decision module is used to c arry the load balancing algorithm proposed in this paper,while other modules are used to assist the path decision module to provide necessary information for it and transfer the generated flow scheduling scheme to the corresponding flow table for distribution.In order to ensure the consistency of the update of the flow table and avoid the phenomenon of packet loss,a conditional classification based sequential flow table update strategy is proposed.