GPTOR: Gridded GA and PSO-based task offloading and ordering in IoT-edge-cloud computing
Arezou Mahmoudi, Leili Farzinvash, Javid Taheri · Results in Engineering · 2025
• Joint task offloading and task-ordering scheme to simultaneously optimize the delay and energy efficiency of exploiting IoT, edge, and cloud computing resources. • Grid the solution space and execute GPTOR in each cell to improve the convergence speed and achieve high-quality solutions. • Customized GA-based solver with a tailor-made crossover operator and two novel mutation operators to distribute tasks among the edge nodes. • Three initialization methods for GA using PSO and RWS to enhance its convergence speed and efficiency in offloading and ordering tasks for execution. Edge computing is a key technology that provides computational resources close to IoT devices. One of the primary challenges in edge computing is determining whether to execute computation-intensive and time-sensitive tasks locally, or to offload them to edge and cloud computing resources, as well as to order them for execution according to their deadlines. Various offloading algorithms have been proposed for these systems, each with its own advantages and disadvantages. Several studies did not exploit all the IoT, edge, and cloud layers, whereas others only considered a few criteria for decision making on task offloading. Other approaches used greedy methods that could not provide high-quality solutions or employed standard optimization algorithms, which took a long time to converge. In this study, we propose an improved genetic algorithm for joint task offloading and ordering to distribute tasks across the IoT, edge, and cloud layers. It includes a novel population initialization scheme that uses various methods, including particle swarm optimization. To increase the convergence speed, the proposed algorithm (GPTOR) splits the solution space into several areas, which is called gridding. The simulation results illustrate that our algorithm outperforms previous schemes by 41.07%, 26.25%, and 28.33% in terms of average delay, monetary cost, and energy consumption, respectively.