Joint Planning for Task Scheduling and Task Result Caching in the Fog Using Reinforcement Learning

The rapid expansion of the Internet of Things (IoT) has greatly increased the number of sensors operating within cloud–fog environments. As more users generate processing tasks, there is a critical need to deliver swift and efficient responses. To address this challenge, it is essential to develop a task scheduling framework that leverages available resources across both fog and cloud domains. This study focuses on the joint optimization of task scheduling and task result caching—a combination that effectively manages computational resources but has been largely overlooked in prior research. While task scheduling has been well explored, result caching has received comparatively little attention, and their integrated application remains underdeveloped. By concurrently employing scheduling and caching strategies, we aim to minimize response times and operational costs. Specifically, frequently requested task results are cached to enable immediate replies to user queries. A key component is the algorithm that decides where to process tasks and where to cache their outputs. We propose a reinforcement‐learning‐based, Actor–Critic real‐time scheduler designed for dynamic, nonepisodic fog–cloud environments with multiple agents. Experimental evaluations demonstrate that our approach significantly outperforms baseline methods (LR‐MMT and LRR‐MMT) and recent state‐of‐the‐art approaches including Digital Twin, PPO‐GRU, A3C‐R2N2, and DDQN in reducing both latency and cost.