Research On Scalability Problems In Software Defined Network

The traditional closed network equipment hinders the development of network innovation technology and increases the difficulty of network optimization.Software-defined network is a network architecture that separates control and forwarding.The control plane provides an open and unified interface through a centralized way,realizes flexible programmable ability,simplifies the management of the network,and improves the flexibility of the network.With the expansion of network scale,SDN has the characteristics of fine-grained flow management and control,separation of control plane and data plane,and centralized control of logic.Therefore,SDN is faced with the scalability problems when it is applied to large-scale networks.How to make use of the advantages of SDN in simplifying network management,avoiding network bottlenecks,and improving the scalability of SDN networks has become the focus of attention of researchers in recent years.The control plane is the core of SDN network and has a vital impact on network performance.This paper analyzes and studies the scalability based on control plane in SDN from two aspects:single controller and multiple controllers.First,based on single controller,the scalability of the control plane is analyzed from four aspects:controller design,optimization of flow table resources,reduction of the number of events handled by the controller,and optimization of routing schemes.For multiple controllers,the scalability issue is analyzed from three aspects:multi-controller design,deployment,and coordination.Furthermore,this paper studies the scalability of a single controller from reducing the number of events processed by the controller and routing schemes.From the flat and hierarchical multi-controller deployment,we study the scalability of multi-controller.The research focuses on four aspects.1.Research on the traffic scheduling mechanism in data center networkCompared with distributed control mode,the centralized control method has more control cost and reaction delay.Especially in large-scale data center networks,the number of network flows are huge.If the flow is scheduled frequently,the load on the controller and the waiting time of the data flows are increased.Therefore,in this section,firstly,the flows with the same attribute characteristics are aggregated.Then,for the aggregated flows,an optimization model is proposed to carry out unified path distribution,and then optimize the load balance of the whole network.For the unaggregated traffic,the path quality is evaluated from the residual bandwidth of the link,the length of the path,and the available bandwidth of the path.Finally,the simulation verification of different flow probability models show that the proposed algorithm has better performance in load balancing,throughput,and link utilization.2.Research on routing optimization algorithm based on segmented routingWith increasing of network size,when SDN performs fine-grained flow management and control,the computational complexity of the control plane increases linearly,and the routing rule space expands,which may lead to a large number of flow table rules cannot be installed in the switch,bringing great pressure to the control plane.In this section we study how to improve the routing optimization performance of the controller in combination with the user's business preferences to improve the controllability of the control plane.In this section,we study routing algorithms based on segmented routing in SDN,and proposes a routing algorithm that combines network performance with user business.First of all,we propose a multi-objective particle swarm method based on the optimization of link weight.Then considering the user's preference and taking the network application service as the guidance,an evaluation hierarchy is established,and the path with the largest evaluation value is selected.The simulation results show that the model of the whole routing algorithm can not only optimize the load balance and link cost of the link,but also evaluate the overall performance of the target link and a single network service in combination with the user service preference,which is conducive to better realization of differentiated service network.It is conducive to the control plane to better realize the differentiated services of the network.3.Research on deployment scheme of flat control planeA single controller cannot meet the management of large-scale WAN,which causes the scale scalability problem,and requires multiple controllers to manage the network cooperatively.The different deployment locations of controllers can affect the response and processing ability of network events.Different controller's deployment locations affect the response and processing ability of network events.In this section,two controller placement algorithms based on density clustering algorithm are proposed for flat multi-controller deployment.The algorithm 1 proposes a method to automatically determine the number of controllers by finding the inflection point of decision graph.In algorithm 2,the number of controllers is automatically determined by kernel density estimation.When selecting the location of the controller and clustering the switch nodes,the transmission delay and communication consumption are taken as the optimization objectives respectively.The simulation results show that the proposed algorithms have better performance in the maximum delay,average delay,maximum consumption and average consumption for intra-controller and inter-controller.4.Research on deployment scheme of hierarchical control planeIn the flat architecture,a consistent synchronization method is required between all controllers to maintain global information,which increases network overhead.The hierarchical architecture logic is relatively simple.The control plane of the hierarchical architecture is divided into several levels.The top controller has a global view and is responsible for making global decisions,while the bottom controller is only responsible for local network control,which can meet the control requirements of large-scale network.In order to improve the scalability of SDN network for hierarchical control architecture,a high-performance hierarchical domain control system is designed in this section.In order to reduce the request delay of the switch and the communication overhead of the control plane,the system is designed in detail from the partition of the underlying switch to the layer of the upper controller.In order to reduce the request delay of the switch and the communication overhead of the control plane,a switch partition algorithm based on the improved Louvain algorithm is proposed.Then a controller deployment algorithm to improve network reliability is proposed.Simulation results show that the proposed scheme has good performance in average request delay,load balancing,and reliability.