On time-based strategies for optimizing flow tables in SDN
Miguel Neves · Lume (Universidade Federal do Rio Grande do Sul) · 2014
Software Defined Networking (SDN) has allowed network operators to greatly improve traffic control through finer-grained network policies. However, finer-grained policies often increase the number of forwarding rules in flow tables, thus aggravating resource scarcity and performance. To tackle current limitations, a promising approach is to use time-based strategies for flow table control. Such strategies attempt to optimize the duration of rules to improve performance. In this work, we characterize the key properties of time-based strategies and discuss their implications on existing SDN networks. In particular, we investigate the effects on the number of rule reinstallations, a key concern that has major influence on flow completion times. Extensive experimentation is performed using representative workloads and accurate parameter sensitivity analysis. Our main findings indicate that strategies proposed in literature are still far from optimal and that their performance depends heavily on a precise parameterization. Additionally, important insights are provided to help understanding the benefits and limitations of current timebased strategies. These insights are useful for designing new forwarding devices, and provide the foundations for novel flow table optimization strategies to be investigated.