Prioritized Constraint Aware Task Offloading Mechanism in Cloud-Fog Computing Using Deep Reinforcement Learning

The rapid emergence of Internet of Things (IoT) applications such as smart cities, intelligent transportation, healthcare, logistics, and wearable systems has increased the demand for scalable and low-latency computing infrastructure. Despite the widespread adoption of traditional cloud computing platforms to handle these applications, offloading all tasks to centralized cloud data centers results in significant latency, high energy consumption, and increased operational costs. Many of these applications are delay-sensitive and computationally intensive, requiring immediate decision-making and localized processing. To address these limitations, we propose a prioritized constraint-aware task offloading mechanism (PCATOM) that efficiently schedules and offloads tasks by considering task-level and resource-level constraints in a fog-cloud computing environment. PCATOM leverages the Deep Deterministic Policy Gradient (DDPG) algorithm, a policy gradient reinforcement learning method designed to balance exploration and exploitation while dynamically learning optimal offloading strategies. The framework reduces overall latency, energy consumption, and execution cost by making intelligent decisions about where and how to offload tasks. PCATOM was implemented using the SimPy simulation framework and evaluated using a combination of statistical workloads and real-world parallel computing traces from NASA and HPC2N. Experimental results show that PCATOM consistently outperforms baseline models such as DQN and A2C, achieving up to 32.5% lower latency, 28.7% lower energy consumption, and 18.4% higher throughput. These results demonstrate the effectiveness and scalability of PCATOM in dynamic and diverse fog-cloud environments.