Abstract:

With the increasing complexity of application components in cloud-native systems, and the majority of them being open-source code, vulnerabilities exploitation in these components has become one of the primary threats to cloud-native security. Moving target defense as an advanced dynamic defense mechanism is widely recognized as an effective solution to this issue. However, the frequent and disorderly configuration transitions in the practical application of moving target defense could reduce system efficiency and service quality, potentially negatively impacting the security of resource-constrained systems. To address the strategy problem of moving target defense in cloud-native stochastic attack-defense environments, this paper combined the modeling advantages of game theory and the solution capabilities of deep reinforcement learning, and proposed a cloud-native moving target defense decision-making method based on stochastic games and the DQN algorithm. The aim was to quickly make optimal moving target defense decision in a large-scale strategy space. The effectiveness and practicality of the proposed model and algorithm are verified through simulation experiments.

Key words: cloud-native, vulnerability exploitation, moving target defense, stochastic game, DQN algorithm