Adaptive Graph Attention Reinforcement Scheduling with Hybrid Optimization for SLA-Aware Multi-Cloud Task Scheduling

Abstract

Task scheduling is one of the most important considerations in a multi-cloud environment, given heterogeneous resources, dynamic workloads, and Service Level Agreement (SLA) requirements. Traditional heuristic and metaheuristic scheduling algorithms cannot achieve optimal performance in large-scale cloud infrastructures because they are not adaptable to changing system conditions. To overcome these constraints, this paper proposes an Adaptive Graph Attention Reinforcement Scheduling (AGARS) system for SLA-conscious multi-cloud task scheduling. The presented framework combines workload prediction using previous temporal transformers, embedding learning via a graph attention network (GAT), and reinforcement learning with the Soft Actor-Critic (SAC) algorithm. Moreover, an Augmented Lagrangian optimization module is also added to meet the required constraints, and a hybrid Harris Hawks Optimization-Differential Evolution (HHO-DE) strategy is used to improve the global search and solution optimization under heavy loads. It compared the proposed AGARS framework to baseline algorithms such as Round Robin, Genetic Algorithm, Particle Swarm Optimization, Deep Reinforcement Learning and GNN-based scheduling. Through experimental findings, AGARS achieves substantial performance improvements, including about a 40% reduction in execution time, a 65% reduction in SLA violations, a 30% increase in throughput, and a 27% decrease in energy usage compared to its counterparts. This set of results an effective and scalable tool for managing intelligent resources in a contemporary multi-cloud computing setup.