Towards Effective Packet Classification

A variety of network security services, such as access control in firewalls and protocol analysis in intrusion detection systems, require the discrimination of packets based on the multiple fields of packet header, which is called Multidimensional Packet Classification. In this paper, we propose a very effective packet classification algorithm called Extended Multidimensional Cuttings, ExCuts in short. As an extension of HyperCuts, which is the best-known existing decision tree algorithm, ExCuts refines the node merging mechanism using a two-step discontiguous space aggregation scheme, which minimizes the number of child nodes. To further reduce the memory usage of the decision tree structure, ExCuts adopts a bit string mapping scheme to compress the large pointer arrays in internal nodes. Due to the significant memory reduction, ExCuts is able to pick a fixed number of cuttings and thus provides explicit worst-case search time. Experimental results show that ExCuts outperforms the best result of existing algorithms on both real-life rulesets and synthetic classifiers