Governing Resource Failures through Reinforcement Learning Scheduling in Fog/Edge Computing: A Review

The exponential growth of IoT applications due to user demands necessitates the emergence of reliable fog/edge computing systems. Because of the wide-open area of these systems, the tendency for failure in communication channels and computing nodes (resources) is unpredictable, especially in real-time IoT applications. Several studies have aimed at tackling fault and failure issues by employing a variety of resource management strategies. Most studies use heuristics or conventional scheduling approaches that leverage the use of reinforcement learning. The element of reinforcement learning aims to support better governance when resource faults and failures occur. This paper provides an analysis of the implementation of reinforcement learning (RL) into fault-tolerant scheduling in fog/edge computing systems. The efficacy of fault-tolerant scheduling is claimed to be improved through RL with regard to system performance. We grouped the discussion based on three themes: resource failures, reinforcement learning (RL) in general, and RL scheduling for fault tolerance. The discussion could further derive solutions for overcoming the challenges of resource failures in fog/edge systems.