SwinTop: Optimizing memory efficiency of packet classification in network devices

Abstract

Packet classification is one of the key functionalities provided by network devices for QoS and network security purposes. Recently the rapid growth of classification ruleset size and ruleset complexity has caused memory performance woes when applying traditional packet classification algorithms. Inheriting the divide-and-conquer idea of pre-partitioning the original rules into several groups for significant reduction of memory overhead, this paper proposes Swin Top, a new ruleset partitioning approach based on swarm intelligent optimization algorithms, to seek for the global optimum grouping of rules. To enhance convergence accuracy and speed up the iterative process, Swin Top employs several novel ideas, such as the introduction of grouping penalty, the combination of PSO and GA, and a new memory usage estimation method. On the publicly available rulesets from Class Bench, SwinTop is shown to achieve 1 to 4 orders of magnitude lower memory consumption than simply applying a traditional packet classification algorithm without ruleset partitioning, and outperform the state-of-the-art partitioning algorithms EffiCuts and ParaSplit on all kinds of large-sized rulesets.