Scalable Packet Classification Using Hybrid and Dynamic Cuttings

Network packet classification is an important functionality provided by modern Internet routers for enabling many network applications such as quality of service, security and differentiated services. Decision-tree based schemes are the most well-known algorithmic solutions for packet classification. But most of them, such as HiCuts and HyperCuts, suffer from memory explosion problem due to uncontrolled rule replications. EffiCuts, a state-of-the-art decision-tree technique, was proposed to reduce memory consumption at the cost of excessive memory accesses. Therefore, these decision-tree based approaches are either memory or performance inefficient, falling short of the needs of high-speed networks. In this paper, we propose HD-Cuts, a scalable decision-tree based packet classification using hybrid and dynamic cuttings to improve storage and performance simultaneously. Using ClassBench, we show that HD-Cuts achieves similarly excellent memory reduction as EffiCuts, but the performance of HD-Cuts is significantly better than EffiCuts in terms of memory accesses. In addition, HD-Cuts is more practical for implementation than EffiCuts.