Supra-linear Packet Processing Performance with Intel Multi-core Processors

This paper describes the performance benefits that can be realized from the utilization of multiple processing cores for communications applications. The paper illustrates the programming concepts of pipelining and flow-pinning in a packet-processing application and presents performance results from the use of these techniques on a popular open source intrusion detection application. Using a real-world network traffic scenario with a high number of TCP connections, the results show supra-linear (more than 6.2X) performance improvements when IP traffic flows are distributed among four cores versus running the same application on a single core.