Performance Evaluation of Software Defined Networks Using Queueing Models

Evolving network requirements have recently made the software defined networking paradigm very popular. In a software defined network (SDN), the data and control functions of network nodes such as routers and switches are separated. In particular, a physically separate controller, which is implemented in software, computes and distributes forwarding tables to routing devices. Such a separation requires an analysis of packet delay performance to evaluate the tradeoffs of using controllers versus a traditional networking architecture. Previous studies have employed simulations and analytical models to evaluate the performance of SDNs before actual deployment. However, these studies are limited to specific topologies, are based on approximations, and cannot be easily extended to more general topologies. The work presented in this thesis employs classed networks of queues to model SDNs. First, a topology that consists of a single switch and a single controller is analyzed using the proposed queueing model. Then, the topology is extended to multiple switches and the methodology is applied to model the extended network. Finally, the classed queueing network model is used to evaluate the deployment of multiple controllers. The single-switch, single-controller topology results are in agreement with previous studies that employ single-class queueing theoretic methods. There is currently limited or no data available to benchmark classed queueing network models of multiple-switch or multiple-controller topologies. Nevertheless, the results give insights into the design and deployment of multiple switches or controllers. For instance, the findings indicate that the power-delay performance is improved when two half-capacity controllers are deployed instead of a single full-capacity controller. In addition, in case there are intermittent controller failures, installing two controllers may be justified for large traffic loads. Keywords: Classed network of queues, OpenFlow, Performance analysis, Queueing theory, Software defined networks