High Performance Flow Feature Extraction with Multi-core Processors

Next generation networks anticipate an increasing amount of network traffic from a wide range of emerging network applications. The features of packet flows (such as the minimal packet inter-arrival time and the number of packets with non-zero options in TCP headers) are used frequently in determining the traffic type and applying security policies. However, the extraction of flow features is difficult due to the increasing line rates, a broad range of network protocols, and a variety of complex flow features. In this paper, we leverage the multi-core processors to speed up the feature extraction process. We design an open source parallel software tool, aiming for processing network packet flows in real-time. We implement the software in four different designs including serial, parallel, pipelined and hybrid architectures. We evaluate the performance of the parallel software tool through measurement experiments. Our experimental results show that each method increases the packet processing throughput by 5-7% in comparison with the previous method. And finally the implementation based on the hybrid architecture improves the packet processing performance by 19.3% than the implementation based on the serial architecture.