Using FlowVisor and evolutionary algorithms to improve the switch migration in SDN

Abstract

The switch migration (SM) issue has been widely addressed in distributed Software Defined Networks (SDN). Most of researches have focused on achieving SM with balanced loads among controllers but little efforts are devoted to equilibrate the service offered to switches when limited resources in control plane are experienced. This paper deals with the problem of dynamic switch migration in multi-controller SDN. Basically, it aims to achieve a trade-off between the whole flow setup delay, the load balancing among controllers and the service fairness among switches by efficiently splitting requests of a switch between multiple controllers. The trade-off is solved using a combination of the Unbalanced Transportation Problem optimization (UTP) and Genetic algorithms (GA). Henceforth, a new genetic algorithm denoted Reference-based Non-Dominated Sorting Algorithm (R-NSGA-II) is proposed to achieve this trade-off. The proposed approach uses the FlowVisor framework as a virtualization layer between switches and controllers. FlowVisor will relay requests coming from switches toward controllers based on current network status with the aim of achieving the above trade-off. The proposed approach is associated with no switch migration cost since assignment is done logically by FlowVisor. Simulation results show that the proposed approach outperforms several existing approaches in terms of the whole flow-setup time, the load balancing among controllers and the request service fairness among switches.