Adaptive Multi-Objective Resource Allocation for Edge-Cloud Workflow Optimization Using Deep Reinforcement Learning

Abstract

This study investigates the transformative impact of smart intelligence, leveraging the Internet of Things and edge-cloud platforms in smart urban development. Smart urban development, by integrating diverse digital technologies, generates substantial data crucial for informed decision-making in disaster management and effective urban well-being. The edge-cloud platform, with its dynamic resource allocation, plays a crucial role in prioritizing tasks, reducing service delivery latency, and ensuring critical operations receive timely computational power, thereby improving urban services. However, the current method has struggled to meet the strict quality of service (QoS) requirements of complex workflow applications. In this study, these shortcomings in edge-cloud are addressed by introducing a multi-objective resource optimization (MORO) scheduler for diverse urban setups. This scheduler, with its emphasis on granular task prioritization and consideration of diverse makespans, costs, and energy constraints, underscores the complexity of the task and the need for a sophisticated solution. The multi-objective makespan–energy optimization is achieved by employing a deep reinforcement learning (DRL) model. The simulation results indicate consistent improvements with average makespan enhancements of 31.6% and 70.09%, average cost reductions of 62.64% and 73.24%, and average energy consumption reductions of 25.02% and 17.77%, respectively, by MORO over-reliability enhancement strategies for workflow scheduling (RESWS) and multi-objective priority workflow scheduling (MOPWS) for SIPHT workflow. Similarly, consistent improvements with average makespan enhancements of 37.98% and 74.44%, average cost reductions of 65.53% and 74.89%, and average energy consumption reductions of 29.52% and 24.73%, respectively, by MORO over RESWS and MOPWS for CyberShake workflow, highlighting the proposed model’s efficiency gains. These findings substantiate the model’s potential to enhance computational efficiency, reduce costs, and improve energy conservation in real-world smart urban scenarios.