High-throughput hash-based online traffic classification engines on FPGA

Abstract

Traffic classification is used to perform important network management tasks such as flow prioritization and traffic shaping/pricing. Machine learning techniques such as the C4.5 algorithm can be used to perform traffic classification with very high levels of accuracy; however, realizing high-performance online traffic classification engine is still challenging. In this paper, we propose a high-throughput architecture for online traffic classification on FPGA. We convert the C4.5 decision-tree into multiple hash tables. We construct a pipelined architecture consisting of multiple processing elements; each hash table is searched in a processing element independently. The throughput is further increased by using multiple pipelines in parallel. To evaluate the performance of our architecture, we implement it on a state-of-the-art FPGA. Post-place-and-route results show that, for a typical 128-leaf decision-tree used for online traffic classification, our classification engine sustains a throughput of 1654 Million Classifications Per Second (MCPS). Our architecture sustains high throughput even if the number of leaves in the decision-tree is scaled up to 1K. Compared to existing online traffic classification engines on various platforms, we achieve at least 3.5× speedup with respect to throughput.