Deceptive VM Deployment Strategy Based on Deep Reinforcement Learning Against Co-Resident Attacks

With the development and popularization of virtualization technology, it offers efficient, flexible services in cloud computing, data centers, etc. However, the sharing of underlying physical devices makes the isolation between different virtual machines (VMs) relatively fragile, leading to new security challenges. Co-resident attack is one of the most representative types. In order to facilitate the construction of side channels to achieve sensitive data theft and other malicious operations, attackers try to co-locate their VMs with target VMs on the same physical device. This paper proposes a VM placement method for virtualization cloud platforms to mitigate the risks posed by co-resident attacks. This study model the VM placement process in the virtualization cloud platform as a Markov Decision Process (MDP) and construct an objective function to minimize co-resident attacks under various constraints. Furthermore, it is believe that after the co-residence occurs, it is equally important to prevent attackers from constructing side channels against the co-resident targets. Therefore, the concept of deceptive defense is introduced to mislead attackers. In addition, this study introduces reinforcement learning and designs a deceptive VM deployment strategy generation method based on PPO (DD-PPO), considering platform security, load balance, and power consumption. Finally, CloudSim is used to build the simulation environment, and the performance of the proposed algorithm is evaluated through experiments. Simulation results show that the proposed algorithm effectively mitigates the threats of co-resident attacks compared to baselines.