Performance Optimization Of Control Plane In Software Defined Data Center Network

Software Defined Networking(SDN)has emerged as a new paradigm that shifts net-work control from distributed protocols to a logically centralized control plane.Through provided open API,it supports flexible network management and rapid deployment of new functionalities.The SDN control plane is typically implemented as a distributed system with a cluster of controllers(e.g.,Onix).Switches are then statically assigned to one or multiple controllers.However,static assignment between switches and controllers results in long and highly varying controller response times,since traffic in data center networks(DCN)fluctuates frequently.Given that dynamic switch migration across controllers is technically feasible as demonstrated by past work,hence,it is critical to apply dynamic assignment to a software defined DCN,for lower controller response time and better utilization of controller resources.Aiming at minimizing the controller response time and control traffic overhead and leveraging M/M/1 queue model,we formulate the dynamic controller assignment(DCA)problem as an optimization,which is validated by testbed experiments.In this problem,each controller has a capacity in terms of the request rate it can manage.The switches are dynamically mapped to different controllers when traffic varies.One key challenge is then to develop an efficient solution algorithm to the DCA problem,so that switches can be timely re-assigned in response to variations of network conditions,even in a large-scale DCN.To this end,based on the key concepts of stable matching and coalitional game theories,we propose a novel two-phase algorithm by casting DCA as a stable matching problem with transfers.We theoretically prove that the proposed two-phase algorithm converges to a Nash stable solution and demonstrate its effectiveness through trace-driven simulations.