Research On Fault Tolerance Mechanism In Software Defined Networks

The architecture of the Internet has seriously hindered the improvement and innovation of the network technology.In order to fundamentally solve the problem,various new network architectures are proposed successively,and the Software Defined Network(SDN)which has the features of separation between the data plane with the control plane,open programmable interfaces,and logical centralized control,is considered as a promising network architecture,and has gained considerable attention from both academia and industry.Network fault is a huge stumbling block for both the Internet and SDN.It occurs occasionally but can result in an immeasurable loss of performance,especially for real-time network services.As a result,fault tolerance becomes an indispensable part for any network architecture.Due to different structures,the failure recovery methods proposed and used in the Internet cannot be applied to SDN directly.Based on the characteristics of SDN,this paper concludes and analyzes the related works on the fault tolerance problem of SDN,and further studies the following four aspects: the architecture of the control plane,the selection of control switches(the switches connecting to the controllers directly),the fault tolerance of the data plane and the fault tolerance of the control plane.Finally,a complete set of fault tolerance mechanisms in SDN is proposed and realized.The main contributions of this paper is summarized as follows:1.Considering the existing control plane architectures can promise a high reliability with the price of a high communication delay,a high construction cost and a high difficulty to realize this paper proposes a new control plane architecture which deploys only two controllers,and decides the amounts and locations of control switches according to three different indicators.Theoretical analyses and simulation results show that our architecture can promise the high reliability and high timeliness.Besides,the proposed architecture is easy to realize and can save deployment cost.2.Considering the existing fault tolerance methods of the data plane for single link failures which adopt the protection mechanism and calculate a backup path for each flow,can promise a fast failure recovery with the price of a high memory consumption,this paper proposes a new method which adopts the protection mechanism,uses LOS to detect failures and calculates a backup path for each link.Theoretical analyses,simulation results and discussion show that our method can promise the high timeliness,strong scalability and wide applicability when recovering from failures.3.Based on the contributions mentioned before,this paper further discusses the fault tolerance methods of the control plane as well as the fault tolerance methods of the data plane for single switch failures and for the multicast scenario.Considering the existing fault tolerance methods need to modify network devices or protocols,or cannot apply to all the topologies,this paper proposes a set of fault tolerance methods which are tailored for different planes(the data plane and the control plane),different data transmission modes(unicast and multicast)and different failure types(link failures and switch failures).Theoretical analyses,simulation results and discussion show that our method can promise the high timeliness and wide applicability when recovering from failures.