An Exploration of Various Quality of Service Mechanisms in an OpenFlow and Software Defined Networking Environment in Terms of Latency Performance

Prevailing network technologies separate control and data planes embedded within networking devices. Problems with this are expensive maintenance of closed-technology network devices and the inability to enforce coherent network-wide policies. Software Defined Networks (SDN) solves these problems with logical centralization based on open standards that use existing networking equipment. Quality of Service (QoS) mechanisms expose configurable parameters that guarantee network performance. These parameters include minimum and maximum rates, number of queues, queue priority, and burst size and are used to implement Class-Based Queuing (CBQ) scheduling that define classes and their scheduling priority. This study is a continuation of a previous effort that explored QoS mechanisms in an SDN environment with the same Class Profiles and Load Configuration[1]. That study focused on bandwidth performance of QoS mechanisms in an OpenFlow SDN environment and included protocol-based metrics. This study explored the latency performance of QoS mechanisms-based CBQ strategies in an OpenFlow implementation of SDN's Southbound Interface. Mininet was used for cost-effective and repeatable emulation of SDN networks. A network topology with a custom controller was deployed on Mininet. Using QoS configurations, CBQ scheduling was implemented for two main Class Profiles: (1) Basic CBQ based on the "Streaming", "Bursty", and "Catch-All" traffic types, and (2) Source CBQ based on Source IP Address groupings. Enforcement points differentiated each Class Profile at the Leaf Switches and at the Core Switch. These Class Profiles and a No CBQ Class Profile were tested for Latency with an averaged Ping scheme at a maximized Load Configuration. Experiment results were based on averaged Ping Round-Trip Times (RTT) measured from all clients to two servers. Results were subjected to Two-Sample T-Tests with Unequal Variance for statistical significance. The previous study[1] showed that CBQs at Leaves' distributed QoS enforcement had significant performance improvement over CBQs at the Core's traditional centralized enforcement with 2% higher bandwidth for Basic CBQ at Leaves and 30% higher bandwidth for Source CBQ at the Leaves. This study further showed that there was a slight but non-significant difference in Latency at 25.08% lower Average RTT for Basic CBQ at the Leaves and 14.41% lower Average RTT for Source CBQ at the Leaves. This indicated that improved bandwidth performance does not come at the price of Latency.