With the rapid development of network, the peer-to-peer (P2P) traffic has become one of the most important traffics on the Internet; meanwhile, it also brings many security problems to the network management. Thus, nowadays, P2P traffic identification is the hottest topic of P2P traffic management. Much effort has been made on this topic, however, effectiveness remains an issue and the classification performance needs to be further improved. Support vector machine (SVM) has advantages with resolving small samples and high dimension for P2P classification problems. However, the performance of SVM is largely dependent on its kernel and parameters. The traditional kernels are hard to map complicated function with high precision and the traditional parameters tuning methods are of low efficiency and difficult to obtain good parameters. As wavelet kernel function is able to approximate a function with high precision and Particle Swarm Optimization algorithm could tune the optimal parameters for SVM. Hence, in the paper, a novel SVM method based on wavelet kernel and particle swarm optimization algorithm (PSO) is proposed for P2P identification. First, the proposed approach tunes the best parameters for SVM with PSO on training data. Subsequently, the wavelet SVM configured with the best parameters is conducted to identify P2P traffic. Experimental results on campus traffic traces indicate that the proposed method is able to identify popular P2P applications with very high accuracy.
A Particle Swarm Optimization Algorithm for Neighbor Selection in Peer-to-Peer Networks
