A Study Of Reliability Assurance In Software Defined Networks

With the construction of digital infrastructure in recent years,the Internet has developed unprecedentedly.As a result,a variety of network applications appear in our vision.At the same time,with the proposal of various new network protocols,the control plane becomes more and more complex.In the traditional network,the control plane is distributed on the network devices.However,with the expansion of the control plane,the network devices become more and more bloated,which hinders the development of the network.Software defined network(SDN)comes into being.It separates the control plane of the network from the network devices,and concentrates it in the controller.The controller is responsible for topology discovery,routing,statistics and resource allocation of the whole network.As a blind device,the network device works according to the guidance of the controller.Due to the separation of control and forwarding,SDN can schedule flow more flexibly and efficiently.So it has been widely concerned by the industry and applied to the network facilities around us.Meanwhile,with the concept of the Internet plus,the connection between network and our life has become more and more close,and the basic necessities of modern human beings depend on the network.Once the network fails,it will greatly affect our work and life,cause immeasurable losses,and even seriously endanger the safety of our lives and property.So it is important to guarantee the reliability of SDN.The control plane and data plane face the challenge of reliability.If the controller fails to distribute the load reasonably when encountering the request burst,it may cause overload and failure of the controller.The existing solution will cause a large synchronization cost of the controller,and it is difficult to deal with a lot of request bursts.When the data plane encounters emergencies like link failure and flow table entry exhaustion on bottleneck forwarding node,it will cause network congestion,routing black hole and other situations,and reduce the QoS.However,the existing solutions either consume a lot of flow entries or have poor timeliness.Therefore,this dissertation is committed to solving the potential risks of both control plane and data plane,so that the network reliability is improved.1.In view of the impact of massive request burst on the control plane,this dissertation proposes the switch controller hybrid mapping and indirect migration solution to prevent the controller from overload and downtime,so as to improve the reliability of the control plane.Large software defined networks use a cluster of distributed controllers to process requests from a massive number of switches.The amount of requests processed by each controller depends on the switches it maps.In this dissertation,the hybrid mapping method is used to map each switch to appropriate controllers.We study a new problem of how to improve the residual capacity available at the controllers to handle request bursts experienced at the switches,in which the residual capacity represents the maximum number of request burst from single switch that the control plane can bear.The prior work either provides poor residual capacity or incurs heavy synchronization overhead.This work proposes a new method called indirect multi-mapping that achieves both high residual capacity and low synchronization cost.We formally define a non-linear integer optimization problem for max-min residual capacity under indirect multi-mapping.We then approximate the problem as two sub-problems:switch-controller mapping selection and weight assignment for each switch-controller mapping.This work solves these sub-problems and formally analyzes their approximate factor.The proposed solution is implemented on an SDN testbed and on large-scale network simulations for experimental studies.Evaluations show that indirect multi-mapping improves the minimum residual capacity by 49.8%on average and reduces the synchronization cost by 41.9-60.3%on average when compared with the alternatives.2.Aiming at the problem of uneven link load caused by insufficient flow table entries of the physical switch,this dissertation proposes virtual switch deployment and load balancing routing solution,which can release the pressure of flow table entries on physical switches by deploying virtual switches,so as to improve the reliability of forwarding nodes on the data plane.It is difficult to provide fully fine-grained flow control for load balancing in an SDN as the flow table size of each SDN switch is usually limited.Inspired by the fact that a virtual switch(vswitch)has more powerful processing capacity and more flow entries compared with a physical switch,this work proposes to achieve the load balancing by incrementally deploying a certain number of vswitches in an SDN.This work formulates the joint optimization of vswitch deployment and routing(JVR)problem as an integer linear program,and proves its NP-hardness.A rounding-based algorithm with bounded approximation factors is proposed to solve JVR.This work implements the proposed algorithm on an SDN testbed for experimental studies and use simulations for the large-scale network.Evaluations show the high efficiency of the proposed algorithm.For example,our algorithm can reduce the link load ratio by about 41.5%compared with ECMP by deploying a small number of vswitches.3.Aiming at the problem of link failures in the network,this dissertation studies the link failure recovery routing with a limited flow table and reliability guarantee,which can quickly recover the network from the impact of link failures,so as to improve the reliability of the link.Flow routing is one of the most important issues in a software defined network,and faces various challenges.For example,each link may not be reliable all the time(or with a failure probability),and the flow-table size on each switch is limited.Existing solutions about reliable flow routing may result in a longer failure recovery delay,a large number of flow entries or massive control overhead.To this end,this dissertation proposes to achieve throughput optimization with the constraint that the forwarding reliability probability of each switch pair should exceed a threshold(e.g.99.9%).This work formally defines the reliable flow routing(PRLF)problem with flow-table size constraint.We present a rounding-based algorithm RDBP for PRLF and analyze its approximation performance.This work implements the RDBP algorithm on the SDN platform and large-scale network simulation.Evaluations show that RDBP can improve the network throughput by about 48.0%and reduce the maximum number of required flow entries by about 53.1%,compared with the existing solutions under the reliability requirement.Through the above methods,this dissertation can properly handle the potential risks in the software defined network,so as to guarantee network reliability and improve the user experience.