Multi-Objective Optimization Framework for Cloud Applications Using AI-Based Surrogate Models

Abstract

The increasing reliance on cloud-based applications presents significant challenges in optimizing resource management while maintaining high levels of Quality of Service (QoS). This paper proposes a multi-objective optimization framework that leverages a deep learning-based surrogate model, specifically a Graph Neural Network (GNN), to balance energy consumption, thermal management, and QoS in dynamic cloud environments. The framework uses the Non-dominated Sorting Genetic Algorithm (NSGA-II) to explore trade-offs between these competing objectives, providing a scalable solution for real-time resource allocation. Evaluation results demonstrate significant improvements, with energy consumption reduced by up to 15%, thermal inefficiencies mitigated by 10%, and SLA violations decreased by 18% compared to baseline models. These findings highlight the effectiveness of the proposed framework in optimizing cloud resource management while maintaining system performance and sustainability. This study paves the way for further advancements in cloud optimization through the integration of AI-driven approaches.