Location-Aware Reinforcement Learning-Based Live Virtual Machine Migration in Fog Computing

One of the most existing problems in live virtual machine migration in a fog environment is the type of devices. The studies assume that all fog nodes in the fog layer are homogeneous and dedicated to fog functionality. However, fog nodes can be implemented over different types of devices (routers, switches, servers, etc.). They are naturally heterogeneous and originally erected to do specific functions. This would create an issue while migrating the virtual machines. Appropriate destinations should be chosen before the migration to avoid any delay in accomplishing the time-critical tasks and to increase the success rate of migrations. Further, fog nodes are location-aware devices. This feature needs to be incorporated while modeling any VM migration solution. To address these challenges, we propose a novel approach to live VM migration in a fog environment that employs Reinforcement Learning (RL). Extensive experiments show that our approach significantly reduces the latency of time- critical applications. Our proposed system, called EVM_MIG, outperforms other models in terms of reduced latency by approximately 50% and this proved that considering device type, function, and location in deciding VM migration in a fog environment can effectively reduce the response time in time- critical applications.