Enhanced Throughput and Accelerated Detection of Network Attacks Using a Membrane Computing Model Implemented on a GPU

Abstract

Membrane computing (MC) is a versatile, nondeterministic, and maximally parallel computing model. We explore the advantages of MC parallelism to flag intrusive connection records in a set of network traffic using a graphic processing unit (GPU) that built on a parallelism platform with a single-program multiple data (SPMD) feature. We build a P system model for attack detection by combining some of the features of a recognizer P system and a tissue-like P system with symport rules. Most previous implementations for intrusion detection have been performed on sequential or minimally low parallel machines called a central processing unit (CPU), so the issue of large data handling has always been a major challenge. Using a massively parallel NVIDIA CUDA architecture, we were able to overcome this problem. Comparison of processing on a GPU and a CPU reveals an increase in average throughput of 50,000 packets/s and more than fivefold acceleration for the detection rate.