Research Of Request Dispatching For Multi-controllers In Software-defined Networking

With the growing size and application of network,the vertically closed traditional architecture has gradually become difficult to adapt to the needs.Software Defined Networking as a new architecture separates control plane from switch plane.Due to its superiority on controllability and flexibility,it is drawing more and more attention.As the control core of SDN network,the scalability and reliability of SDN controller play a decisive role in the performance and stability of network services.To improve scalability and avoid a single point of failure,the control plane is typically implemented as a distributed system with a cluster of controllers,Switches are then statically assigned to control plane without service distinction in these systems.However,static assignment between switches and controllers results in long and highly varying controller response times because requests fluctuate frequently and cause hot spots among some controllers.The assignment without service distinction leads to service performance damage because different type of services have significant gap in execution time and need multiple levels of protection.Towards this,we present an approach that solves the problem of dynamic request dispatching while providing per-service performance guarantees.We introduce the architecture of RT-Scheduler which helps in service protection.Based on that,we consider dynamic request dispatching problem so as to minimize the processing delay of services.We formulate the problem as the combination of stable matching and cooperative game to optimize consecutively.Theoretical analysis is given to our approach based on approximation and complexity.The evaluation results show that our approach rapidly computes the near-optimal solution.In addition,we design and implement a proto-system SDN-RT,constituted by the orchestration,the resource manager and RT-Scheduler.Organizers can efficiently and practically manage network services by it.The evaluation results show that our approach rapidly computes the near-optimal solution.Compared to the baseline strategy,our approach has almost twice as much acceptance rate for requests and decreases at most 75% service delay.The results of Mininet simulation show that our RT-Scheduler has fast convergence speed and tolerable extra overhead in a large-scale real-time scenario.