A Novel Artificial Intelligence Based Dynamic Task Scheduling and Load Awareness

Abstract Fog computing has emerged as a solution to the latency and bandwidth constraints of cloud-centric architecture yet efficiently scheduling heterogeneous IoT workloads while minimizing energy use and Service Level Agreement (SLA) violations remain challenging. The proposed framework integrates four complementary algorithms to enhance dynamic task scheduling in fog computing environments. First, a task classification module distinguishes IoT workloads as resource-intensive or non-resource-intensive based on task length, resource requirements, user preferences, and delay constraints. Second, a task prioritization module assigns tasks to major, middle, or minor priority levels using an emergency degree metric derived from waiting time, time-to-live, deadlines, and execution time. Third, the SUPERior-net-aided bipartite Deep Q-Network (SUPER DQNET) employs deep reinforcement learning to schedule non-resource-intensive tasks across fog nodes, optimizing the trade-off between energy consumption and latency. Finally, the Boosted Binary Owl Optimization (BOON) algorithm performs optimal virtual machine (VM) selection for task forwarding by jointly considering CPU utilization, bandwidth, queue length, system load, response time, makespan, energy usage, and SLA compliance. Implemented in iFogSim2, the proposed framework was evaluated against state-of-the-art methods including EEIOMT and EaDO, as well as baseline scheduling strategies such as Random Scheduler, Round Robin (RR), and First-Come, First-Served (FCFS). Experimental results show that our approach achieves up to 35\% reduction in energy consumption, about 40\% improvement in throughput, and over 40\% reduction in SLA violation time, while maintaining low latency. These findings demonstrate the effectiveness of the proposed solution in delivering adaptive, energy-efficient, and SLA-compliant task scheduling for next-generation IoT applications in fog computing environments.