Towards transforming OpenFlow rulesets to fit fixed-function pipelines

Abstract

OpenFlow feature support differs between devices due to devicespecific hardware constraints. OpenFlow places the burden of addressing these differences on the controller, which increases development cost and restricts device interoperability. This paper investigates reducing this burden by algorithmically transforming an existing ruleset to fit an incompatible fixed-function pipeline to improve device interoperability. Existing rule-fitting schemes in the literature require metadata to link rules between tables into a path through the pipeline, but not all pipelines support metadata. We developed a novel approach that does not rely on any particular pipeline features, like metadata, and considers the pipeline's constraints, including both the matches and actions available. This paper presents our implementation, including ruleset preprocessing techniques, methods of transforming rules, and how we use a partially constrained boolean satisfiability problem to select from these transformations and build the final ruleset. While future work remains towards real-world deployment, our approach demonstrates fitting rulesets to fixed-function pipelines without metadata is feasible, and our techniques to reduce the size of the problem are beneficial.