On the resource trade-off of flow update in software-defined networks

In software-defined networks, packet forwarding is performed by installing rules on the switches' flow tables. After the rules are installed, the controller needs to dynamically update the rules during runtime for a variety of reasons including traffic engineering, policy changes, network maintenance, and so on. A single forwarding policy update often consists of rule modifications on multiple switches simultaneously. Since the update process requires moving flows to different paths in a consistent and correct manner, multiple steps are usually involved. Thus, additional resources are needed to handle the flow update to ensure correctness and performance, such as extra bandwidth and flow table entries. In this work, we analyze different existing mechanisms of flow update from their resource utilization perspectives. Specifically, we study the impact of bandwidth and flow table size on the performance of flow update, and their interactions. We provide both qualitative analysis of the tradeoff between these two kinds of resources, and the quantitative simulation results of this trade-off under different realistic network topologies. Our observation is important to the problems related to flow update, based on which we further illustrate its usefulness with three applications.