Deep Reinforcement Learning-based Network Moving Target Defense in DPDK.

Aiming at the problems of difficult deployment, high latency, and serious performance loss of current network moving target defense, this paper designs a network moving target defense system based on deep reinforcement learning by using Data Plane Development Kit (DPDK). In the paper, we first give the framework of the network moving target defense system based on DPDK. On this basis, in order to reduce the performance loss caused by shuffling network attributes, we design a performance model based on queuing theory and security model based on probability. In addition, we construct the relationship between performance and security by increasing the weighting factor, and then design the shuffling policy selection method based on the Proximal Policy Optimization algorithm. Simulation experimental results show that the optimal shuffling period obtained by the proposed algorithm can effectively enhance the security of the system and reduce the performance loss of the system. This method improves performance and security by 96% and 60%, respectively, compared to the comparison methods. In addition, the real environment verifies that the system can enhance the anti-scanning ability of the system while ensuring stable data transmission.