An exploration of various quality of service mechanisms in an OpenFlow and software defined networking environment

Technology prevalent in networks today logically detaches the control from the data plane yet physically deploys them on the same device on all network devices. This has negatively resulted in expensive management of disparate and closed network devices and the inability to implement coherent, synchronized, and wide-ranged network policies. Software Defined Networking (SDN) solves these issues by providing an open and virtually centralized network system through cost-effective and gradual network introduction that minimizes investment in infrastructure. The OpenFlow protocol defines communications from the control plane where the logical center of the network, the controller, resides to the packet-routing devices in the data plane and vice-versa. The Mininet SDN emulator mimics real-world SDN network environments for rapid system testing and proof-of-concept experimentation. Quality Of Service (QoS) represents a ripe and proven technology that allows for the enforcement of priorities and assured performance levels on network links and objects. This study examined the application of QoS-based methods to an SDN OpenFlow environment. QoS parameters were utilized to implement rudimentary Class-Based Queuing (CBQ) scheduling algorithms on a custom OpenFlow controller within a Mininet-emulated SDN network topology. Two CBQ algorithms were used in the experiments contrasted by the classes on which they were based. One algorithm is Basic CBQ which modeled scheduling based on the “Streaming”, “Bursty”, and “Catch-All” traffic types. The other algorithm, Source CBQ, was based on predefined Source-IP address groupings of client hosts. Each Class Profile was broken down in terms of QoS enforcement points at the Leaf Switches and at the Core Switch. Tests were conducted on the topology with these CBQ-based Class Profiles to gather performance data and observe any issues encountered. Experiment results showed that CBQs at the Leaves yielded better overall average bandwidth than CBQs at the Core with Basic CBQ at the Leaves at 2% higher than Basic CBQ at the Core and Source CBQ at the Leaves at 30% higher than Source CBQ at the Core. This indicated that QoS implemented at the Client Leaf Switches rather than at the Core Switch yielded better performance as they were not prone to the same bottlenecks as the Core Switch. Additionally, experiment results seemingly showed that Basic CBQ, whose classes are based on traffic types, yielded optimum performance at less variability as it managed bandwidth better by segmenting traffic and limiting errors.