Parallel packet classification using GPU co-processors

Abstract

In the domain of network security, packet filtering for classification purposes is of significant interest. Packet classification provides a mechanism for understanding the composition of packet streams arriving at distinct network interfaces, and is useful in diagnosing threats and uncovering vulnerabilities so as to maximise data integrity and system security. Traditional packet classifiers, such as PCAP, have utilised Control Flow Graphs (CFGs) in representing filter sets, due to both their amenability to optimisation, and their inherent structural applicability to the metaphor of decision-based classification. Unfortunately, CFGs do not map well to cooperative processing implementations, and single-threaded CPU-based implementations have proven too slow for real-time classification against multiple arbitrary filters on next generation networks. In this paper, we consider a novel multithreaded classification algorithm, optimised for execution on GPU co-processors, intended to accelerate classification throughput and maximise processing efficiency in a highly parallel execution context.