Research On Multi-controller Placement Strategy For Latency And Load Optimization In Software Defined Networks

Software Defined Networks(SDN)strips the control logic from the data layer and forms the control layer separately,enabling flexible management of the network.However,as the increasing scale of the network,the single-controller deployment solution cannot manage the entire network due to self-defects.Therefore,multi-controller placement problem was raised,that was,how to select the number and location of controllers and the mapping relationship between switches.Previous studies have shown that the number and location of controllers can greatly affect the performance of network latency and load.In this thesis,we study the placement strategy of multi-controllers for latency and load optimization in software-defined networks.The research contents are as follows:(1)Multi-controller Static placement strategy for latency and load optimization: This strategy comprehensively considers the effects of network propagation latency,controller load and queuing latency on multi-controller static placement issues,establishes corresponding models and proposes load balancing Algorithm(BCRA)and Genetic Algorithm(GA).The load balancing algorithm(BCRA)first determines the number k of controllers,and selects the node with the highest degree to construct k spanning trees for the root node of the controller to generate the initial k low-latency and load balanced subnets;In GA,firstly,the greedy thought is used to determine which set of controllers will be selected,and then they will be coded.Through constant order selection,crossover and mutation,an optimized multi-controller placement scheme is finally obtained.Experiments show that,compared with k-center and k-means algorithms,BCRA and GA have better network load balancing performance in small and medium-sized networks,while in large-scale networks,GA algorithm has propagation latency and queuing latency.And the performance of load and other aspects is better than BCRA,which makes the load balancing rate increase by 49.7% on average,ensuring low propagation latency of flow requests,queuing latency and controller propagation latency.(2)Multi-controller dynamic placement strategy for latency and load optimization: This strategy comprehensively considers the impact of network queuing latency,switch migration costs,and controller load on dynamic multi-controller placement issues.This strategy further discusses the problem of controller overload due to changes in traffic based on the results of static placement strategies.The problem is mainly divided into two cases: 1)when the controller is overloaded,the network still can be operated in a normal way due to the sufficient load of controller.This strategy proposes the PODA algorithm,which is intended to readjust the control domain of each controller while the switch migration cost is relatively low,it can balance the load between the controllers and reduce the network queuing delay;2)In the case that the controller and entire network both are overloaded and the current number of controllers can no longer operate properly.Here we propose the FODA algorithm,calculate and select the number and location of newly added controllers,and finally makes the network load balanced.Experimental results show that the multi-controller dynamic placement algorithm proposed in this chapter can effectively balance the load between controllers,ensure less switch migration costs and lower network queuing latency at the same time,and can be deployed in largescale networks.