Energy-aware load balancing in cloud environments using graph neural networks and grey wolf optimization

Cloud computing infrastructures must constantly maintain a balance between resource utilization and energy consumption to work effective under dynamic workloads. The training expense and generalizability of machine learning-based load balancing algorithms are both considerable, while existing methods often fails in adjusting in real time conditions. To work with both energy-aware and scalable load balancing, this research proposes a hybrid framework that combines Graph Neural Networks (GNNs) with the Grey Wolf Optimization (GWO) method. GNN gives accurate workload representation, representing the intricate structural relationships among virtual machines (VMs), tasks, and resources. Afterwards, the usage of GWO maximizes task-to-virtual machine mappings by reducing load imbalance, energy consumption, and task completion time. The hybrid framework utilizes sustainability parameters while constantly adapting to workload variations, unlike traditional methods. The experiments are done in CloudSim, where the proposed hybrid model reduced energy consumption by approximately 18-27%, decreases task completion time by 12-20%, and improves resource utilization balance by 15-22% when compared to existing approaches. The simulations are run in CloudSim Plus environment using real-world traces.