Efficient Load Balancing Adjustment for Reliable Multiple Controllers in SDN Networks

碩士 === 國立中興大學 === 資訊科學與工程學系 === 106 === By separating control plane from data plane, Software-Defined Networks (SDNs) are flexible in network traffic routing determination. As the network’s coverage increases, a single SDN controller could become a severe bottleneck. Equipping with multiple controll...

Full description

Bibliographic Details
Main Authors: Kai-Yu Wang, 王剴佑
Other Authors: 高勝助
Format: Others
Language:en_US
Published: 2018
Online Access:http://ndltd.ncl.edu.tw/handle/755zea
Description
Summary:碩士 === 國立中興大學 === 資訊科學與工程學系 === 106 === By separating control plane from data plane, Software-Defined Networks (SDNs) are flexible in network traffic routing determination. As the network’s coverage increases, a single SDN controller could become a severe bottleneck. Equipping with multiple controllers in the SDN environment could be a feasible solution to the scalability problem. In this thesis, we focus on ensuring balanced load among all controllers for the purpose of better system performance. The proposed mechanism works on each controller through the three logical components: Load Collector, Load Balancer, and Switch Migrater. Since all controllers get involved in the determination of load balancing, they should keep the global loading information. A load threshold (T) of the controller is pre-defined and dynamically adjusted. The interval from 0 to T is further divided into 4 sub-intervals to reduce the bandwidth use on information transmission. When the new and the previous loads are in different intervals, an announcement of the new load is delivered to all other controller. Occasionally, the load balancer examines the loads among all controllers and determines whether there is a necessity of migrating switches to other controller. Once a request is issued, the switch migrator sends out the forwarding rule to the switches and notifies the corresponding controller for the new association. The experimental environment consists of ten inter-connected ONOS controllers and each controller is associated with three switches. By using Cbench to generate traffic to simulate various loads of each switch, our proposal is able to adjust the loading of the ten controllers to meet the load-balancing requirement. Moreover, by comparing with the standard deviation (SD) of traffic load on all controller before load balance and after, the load of the controllers after we do balance is closer to the average load than before. Hence, our system utilization is improved effectively.