Machine learning-based multi-objective optimization technique for load balancing in integrated fog-cloud environment
Niva Tripathy, Subhranshu Sekhar Tripathy, Satyananda Champati Rai · 2024
With the rapid proliferation of Internet of Things (IoT) devices and edge applications, the need for efficient load distribution across fog nodes and cloud servers becomes critical to achieve low-latency, real-time processing, and optimal resource utilization. The integration of fog and cloud computing has emerged as a promising architecture to address the growing demands of resource-intensive applications and services in the era of the IoT and edge computing. The load optimization problem is a multi-objective task, considering objectives such as minimizing response time, energy consumption reduction, maximizing resource utilization, and maintaining load balancing. To support the machine learning model, extensive data is collected from fog nodes, cloud servers, and the underlying network infrastructure. This data encompasses historical workload patterns, node capacities, processing times, network latency, and energy consumption metrics. The collected data is preprocessed and engineered into relevant features for training the machine learning model. This chapter explores a range of machine learning algorithms suitable for handling multi-objective optimization tasks, including multi-objective evolutionary algorithms, reinforcement learning, and ensemble methods. These algorithms are compared and evaluated based on performance metrics such as Pareto optimality, convergence, and computational efficiency. The proposed technique enables the integrated fog-cloud system to dynamically allocate computational tasks and workloads across fog nodes and cloud servers, considering the diverse objectives simultaneously. Through extensive simulations and experiments, the efficacy of the machine learning-based approach is demonstrated, showcasing its ability to adapt and optimize load allocation in real-time, resource-constrained environments.