Defense Mechanism of Interest Flooding Attack Based on Deep Reinforcement Learning
Jie Zhou, Jiangtao Luo, Lianglang Deng, Junxia Wang · 2020
At present, the Internet is facing technical challenges in terms of dynamic, security, and scalability, and its development speed has been difficult to adapt to the explosive expansion of the global network scale. Because of its excellent characteristics, named data networking has been widely concerned by people, and has become the representative architecture of the future network. However, because of the characteristics of stateful forwarding, a new attack method, interest flooding attack, is introduced. The attacker makes a large number of requests to send malicious requests, thus exhausting the PIT table resources in the intermediate router. The interest packet is not cleared until it is out of date, so normal requests cannot be processed.To solve the above problems, we propose a defense scheme of interest flooding attack based on deep reinforcement learning. The agent is trained by collecting the relevant data of some routing nodes. Appropriate action space is designed to enable agents to take different defense measures according to different network states to achieve the purpose of defending IFA. Finally, the PIT entry is introduced into the reward function to evaluate the actions made by the agent. Finally, we compare the number of received packets, the number of retransmissions of interest packets, and the average request delay of users, and we find that the proposed mechanism can better resist the flooding attacks of interest packets.