Self-Organizing Maps-Based Flexible and High-Speed Packet Classification in Software Defined Networking

This work studies the application of growing hierarchical self-organization map (GH-SOM) for high performance and flexible packet classification in the context of software-defined networking (SDN). Highly flexible packet classification is necessary for SDN since SDN applications enable fine-grained policies (i.e., increased number of rules) and install packet flow classification rules on-the-fly during flow setup. We show that a hierarchical tree of SOM is fast, flexible, and retraining of SOM-tree is not required when only a small number of rules are updated. Therefore, SOM-tree adeptly absorbs updates in rulesets facilitating a flexible packet classification, a key requirement of SDN. Our results for rulesets generated using ClassBench [1] shows high accuracy in packet classification. Classification accuracy is also characterized when network rules are updated on-fly