Neural Network with Evolutionary Algorithm for Packet Matching

The performances of network forwarding device are determined by the efficiency of packet matching algorithm. It is difficult for network device based on traditional algorithm acting as software core to achieve linear forwarding. This paper proposes a new packet matching algorithm, which achieves packet matching by combination of evolutionary algorithms and neural networks. Firstly, the evolutionary algorithm is used to evolve weight value and activation function of neural networks. Secondly, influence factor is applied to prune the neurons of the hidden layer. And finally, back-propagation algorithm is utilized to fine-tune the neural network. So a compact and efficient neural network structure to solve packet matching is created by the creative procedure. Data experiments show that this new algorithm effectively improves the performance of packet matching compared with the classical algorithms. And it can completely solve the problem of large-scale rule packet matching.