Research On The Key Technology Of The Load Optimization For Multiple Controllers In Software Defined Networks

With the high-speed development of the Internet, the scale of the user and the quantity of the infrastructure have been expanding for a further step. The static structure and managing strategy of traditional network have been difficult to match the trend of the dynamic development of various applications. Recent studies focus more on Software Defined Networks (SDN), a new form of network structure, for its advantages like programmability and easy to manage, as it solves the device complexity and configuration problems in traditional networks.The typical SDN with single controller realize the construction of the network while bringing some inherent shortcomings at the same time:one-node failure and scalability. Therefore, a variety of multiple-controller structures have been proposed. However, the mapping between controller and switch in most structures is static, which cannot fit the change of network traffic, leading to load imbalance among controllers. As a result, the dynamic structure comes. It solves the problem of load imbalance through two methods. The first method is to reassign switches within overloaded controllers to the other controllers dynamically. The second one is based on the first approach. It designs the control layer through adding or deleting controllers to achieve load balance of controllers. But the two methods have the following problems:(1) extensive switch migration mechanism; (2) the inefficiency of the switch migration; (3) the existed mechanisms never focus on load balance part.Consequently, with the consideration of load balance, this paper focuses the extensity and low efficiency of the exited migration mechanisms. First, this paper proposes a switch migration strategy based on immune particle swarm optimization algorithm to improve the extensity of switch migration. Second, a mechanism of switches migration based on progressive auction (PASMM) is proposed to improve the inefficiency. Finally, in order to adapt the dynamic change of traffic in actual networks and improve the resource utilization, this paper proposes a controller dynamic adjustment algorithm. The details are as follows:1. A strategy of switch migration based on particle swarm optimization with immune algorithm(SMPSO-IA)It models the controller-switch mapping as 0-1 programming problem, and encodes the controller-switch mapping as the particle position vector. The fitness of particles is defined by the mean square error of controller resource utilization after d iterations, then the optimal deployment of switches is the optimal position searched by SMPSO-IA. Simulation results show that, compared with typical algorithms, the algorithm achieves better load balance among controllers, reducing the response time of the PACKET_IN messages, improving the system response speed and bringing high effectiveness.2. Progressive Auction based Switches Migration Mechanism(PASMM)In the mechanism, the controllers which have light load are auctioneers and sell their remaining resources. The migrated switches are bidders and bid for the resources of controllers to get services. By increasing the trading price of the over-demanded controllers’resources, PASMM completes the auction and redeploys the controllers and switches. The step size of price is limited to reduce the auction time. In order to minimize the impact caused by the switches migration, we not only illustrate how to select the switches which will be migrated from the over-load controllers, but also present how to implement the migration in software defined networks. Simulation results show that the mechanism can quickly reach convergence and load balance among controllers.3. Controller Dynamic Adjustment AlgorithmBy dynamically turning on or off the controllers, the algorithm meets the needs of the network dynamic changing traffic. On the basis of guaranteeing transmission delay, it balances the load among controllers. The algorithm has two stages. The first stage models the selection of the controllers by 0-1 knapsack problem, then the greedy algorithm is used to solve the problem and the deployment of controllers and switches which meet the capacity constraints is output. The second stage uses the particle swarm optimization based switch migration mechanism we proposed, then the global load balance deployment is achieved. Simulation results show that the algorithm can effectively reduce the number of controllers and improve the utilization of controllers’ resources, realizing the load balance among controllers.