A performance evaluation of Software Defined Networking load balancers implementations

Abstract

Traditional network load balancers have been developed and deployed by combining software and hardware into network appliances. This approach makes network load balancers inflexible (limited number schedule disciplines implemented), expensive and hard to perform network management tasks. Software Defined Networking (SDN) is emerging as new network paradigm that separates software and hardware roles in devices (vertical integration), enables programmability of networks, generalizes network devices and functions, and centralizes network management. Those ideas can be apply to development of new load balancers to overlap the disadvantages of traditional load balancers. Thus, this work evaluated the performance of SDN load balancers implementations with different schedule disciplines (Random, Round-Robin and CPU Usage), and the reactive and active SDN load balancers. Through careful performance evaluation based on collected data and the reviewed literature, we concluded that SDN load balancers implementations are very effective, viable, flexible and inexpensive solutions to distributed the network load.