Authors: Om Sawant, Gunjan Shahade, Atul Sanap, Shailesh Pawar, Madhuri Shinde
Abstract: The widespread adoption of Internet of Things (IoT) systems has resulted in a large volume of time-sensitive data that requires fast and efficient processing. Although cloud platforms provide extensive computational capabilities, the physical separation between data-producing devices and remote cloud infrastructures frequently introduces noticeable delays, jitter, and bandwidth inefficiencies. Fog computing addresses these shortcomings by relocating processing tasks toward the network’s periphery; however, the decentralized and heterogeneous composition of fog resources complicates the design of effective scheduling strategies. Recent progress in Artificial Intelligence (AI), especially in the field of Deep Reinforcement Learning (DRL), have enabled adaptive and context-aware scheduling solutions capable of responding to dynamic changes in fog–cloud systems. This study presents an in-depth examination of AI-oriented scheduling mechanisms for fog computing, with emphasis on system design principles, algorithmic trends, and comparative performance outcomes. Conventional scheduling heuristics, machine-learning-based methods, and contemporary DRL approaches—including multi-agent and multi-objective frameworks—are critically analyzed. The review also identifies persistent challenges related to scalability, mobility, resource constraints, and security-aware decision-making. Overall, the findings demonstrate that AI-driven scheduling enhances responsiveness, load distribution, and resource utilization in emerging fog-supported IoT environments.
DOI: http://doi.org/10.5281/zenodo.20769734
