Network traffic on-line classification using decision tree fast parallel processing strategy

At present, most of the current traffic classification technology is adopted by the means of offline classification, unable to realize real-time online classification. A novel approach for network traffic online classification is proposed in this paper. A strategy of converting the C4.5 decision tree data structure to parallel data structure called encoded data structure is proposed. The encoded data structure is very easy for hardware realization, based on FPGA parallel and pipeline technology. It only needs two clock cycles to complete C4.5 decision tree search process without extra write and read controls. Experimental results show that the throughput of the classification system can reach to 1 Gbit/s on the NetFPGA2.1.3 platform, and we expect that this design can be expanded to the NetFPGA-10G platform and the throughput can reach to 10 Gbit/s. The classification accuracy can reach to more than 97% if we choose the appropriate evaluation algorithm and search algorithm to obtain the effective features set.