OpenFlow Accelerator: A Decomposition-Based Hashing Approach for Flow Processing

To support scalable, flexible software-defined networking, OpenFlow is designed to provide granular traffic control across multiple vendor´s network devices for efficient flow processing. Decision-tree packet classification algorithms do not scale to the number of flow table fields while decomposition algorithms such as RFC fail to provide necessary incremental update and determinism. Since searching in a single field is well studied, e.g. Longest Prefix Match (LPM) for prefix fields, we propose a decomposition approach which performs individual search on each flow table field, aggregates these results and conducts a query in a single hash table. Our approach scales well to the number of fields and allows incremental update. Meanwhile deterministic query is enabled for high-speed search. As far as we know our proposal is the first efficient decomposition approach to address multidimensional match in an OpenFlow flow table with an arbitrary number of fields as well as any match type. Theoretical analysis and experiments using synthetic classifiers justify the performance improvement.