The Internet of Things (IoT) has become a pervasive phenomenon, with applications in a multitude of sectors, including healthcare, smart agriculture, smart cities, transportation, and water management. This has led to a significant generation of Big Data. In order to process this substantial volume of data efficiently, there is a pressing need for a platform capable of handling large quantities. However, real-time applications face challenges in cloud processing due to high latency. As a complementary infrastructure to the cloud, fog computing emerges as a viable solution by facilitating task processing, networking, and data storage in cloud data centers accessible to mobile users. The offloading of tasks represents a promising solution to the resource constraints inherent in IoT applications, particularly within the context of fog computing. This process entails the execution of particular components of mobile applications within a fog–cloud environment, to reduce execution time and energy consumption. The objective of our research is to optimize task offloading in IoT within heterogeneous environments, taking into account conflicting constraints. This optimization challenge is formulated as a multi-objective problem, with a particular focus on energy consumption and latency, as well as quality of service metrics. The proposed solution, TOF-NSGAII, is designed to respect the finite resources of fog computing, balancing workloads to meet the latency requirements of IoT tasks. The widely employed meta-heuristic, the non-dominated sorting genetic algorithm (NSGA-II), has been adapted to generate a set of non-dominated multi-objective task offloading optimization solutions, considering both energy consumption and latency. The experimental results demonstrate the efficacy of TOF-NSGAII in generating task offloading solutions that distribute executed tasks between fog and cloud computing environments in a judicious manner, based on their specific requirements. Furthermore, the generated non-dominated solutions demonstrate optimality in terms of energy consumption, with an average reduction of 12.18% compared to alternative approaches. It is noteworthy that our approach introduces only a marginal increase in latency, amounting to 0.38%, which can be considered negligible.
