Deep Reinforcement Learning-Based VM Migration for IoT Security
Jagruti Sahoo · 2024
Internet of Things (IoT) has become a popular technology due to its widespread applications in smart cities, smart homes, and smart farms. Recently, fog computing-based infrastructures have been used to deploy IoT networks, where resource nodes, called fog nodes host the IoT applications residing in virtual machines (VMs). However, VMs can be targets of cyber-attacks resulting in the disruptions of IoT applications. Hackers can reach the VMs by first compromising the fog nodes and then exploiting virtualization-based exploits. In this paper, we propose a moving target defense (MTD) scheme to ensure the resiliency of VMs. The MTD scheme is designed based on VM migration, where the VMs are migrated among the fog nodes to hide the physical location of VMs from hackers. The MTD scheme is designed by modeling a Deep Reinforcement Learning (DRL) agent that interacts with the fog infrastructure by taking migration actions and receiving rewards. We designed the reward by considering the reconnaissance activities at the fog nodes. As the amount of reconnaissance activities provides a measure of the likelihood of an attack against fog nodes, it allows the agent to optimize its actions i.e., actions that can avoid unreliable fog nodes. We performed multiple tests to assess the performance of our agent. Our results show that the agent consistently performed well in all test scenarios by selecting a reliable fog node as the migration destination, thereby ensuring continuous operation of the VM and the associated IoT application.