Scholarlink

<p>Efficient scheduling of aperiodic tasks in fog-cloud environments poses significant challenges due to dynamic workloads, heterogeneous resources, and the need to simultaneously optimize for deadline adherence, energy efficiency, scalability, and Quality of Service (QoS). Traditional scheduling approaches, including heuristic methods such as Earliest Deadline First, and metaheuristic techniques such as Genetic Algorithm and Particle Swarm Optimization, are often limited by inflexible prioritization and poor adaptability to workload variation. Similarly, single-agent reinforcement learning models such as Proximal Policy Optimization and Asynchronous Advantage Actor-Critic exhibit limited scalability and delayed convergence in distributed settings. To address these limitations, this paper proposes fog-adaptive multi-agent scheduling optimization (FAMASO), a deep multi-agent reinforcement learning framework that integrates Earliest Deadline First-based task prioritization with Proximal Policy Optimization and Recurrent Neural Networks for real-time, decentralized scheduling in fog-cloud infrastructures. FAMASO dynamically adapts to system conditions, learns temporal task patterns, and supports distributed decision-making to optimize key objectives. Experimental evaluations across workloads of 100 to 900 aperiodic tasks and fog networks ranging from 10 to 100 nodes show that FAMASO consistently outperforms baseline algorithms, including Earliest Deadline First, Genetic Algorithm, Particle Swarm Optimization, Multi-Agent Deep Deterministic Policy Gradient, Asynchronous Advantage Actor-Critic, and Proximal Policy Optimization. FAMASO achieves up to 95.66% deadline adherence, 38% energy savings, 56% makespan reduction, 7.8% improvement in resource utilization, 70% reduction in scheduler time, and notably higher throughput, completing more tasks per unit time than all compared methods. Additionally, it achieves SLA violation rate reductions of up to 80%, demonstrating strong compliance with real-time constraints. These results validate FAMASO’s effectiveness, scalability, and practical potential for deploying efficient, SLA-aware task scheduling in next-generation fog-cloud computing environments supporting time-sensitive IoT applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

FAMASO: fog-adaptive multi-agent scheduling optimization

  • 期刊论文

  • 发表日期：

    2025年8月19日

查看全文
  • Ganesan Nagabushnam,
  • Kyong Hoon Kim,
  • Yundo Choi

摘要

Efficient scheduling of aperiodic tasks in fog-cloud environments poses significant challenges due to dynamic workloads, heterogeneous resources, and the need to simultaneously optimize for deadline adherence, energy efficiency, scalability, and Quality of Service (QoS). Traditional scheduling approaches, including heuristic methods such as Earliest Deadline First, and metaheuristic techniques such as Genetic Algorithm and Particle Swarm Optimization, are often limited by inflexible prioritization and poor adaptability to workload variation. Similarly, single-agent reinforcement learning models such as Proximal Policy Optimization and Asynchronous Advantage Actor-Critic exhibit limited scalability and delayed convergence in distributed settings. To address these limitations, this paper proposes fog-adaptive multi-agent scheduling optimization (FAMASO), a deep multi-agent reinforcement learning framework that integrates Earliest Deadline First-based task prioritization with Proximal Policy Optimization and Recurrent Neural Networks for real-time, decentralized scheduling in fog-cloud infrastructures. FAMASO dynamically adapts to system conditions, learns temporal task patterns, and supports distributed decision-making to optimize key objectives. Experimental evaluations across workloads of 100 to 900 aperiodic tasks and fog networks ranging from 10 to 100 nodes show that FAMASO consistently outperforms baseline algorithms, including Earliest Deadline First, Genetic Algorithm, Particle Swarm Optimization, Multi-Agent Deep Deterministic Policy Gradient, Asynchronous Advantage Actor-Critic, and Proximal Policy Optimization. FAMASO achieves up to 95.66% deadline adherence, 38% energy savings, 56% makespan reduction, 7.8% improvement in resource utilization, 70% reduction in scheduler time, and notably higher throughput, completing more tasks per unit time than all compared methods. Additionally, it achieves SLA violation rate reductions of up to 80%, demonstrating strong compliance with real-time constraints. These results validate FAMASO’s effectiveness, scalability, and practical potential for deploying efficient, SLA-aware task scheduling in next-generation fog-cloud computing environments supporting time-sensitive IoT applications.