Scholarlink

<p>Fog computing has become essential for real-time applications in domains such as autonomous driving, smart cities, Industry 4.0, and large-scale IoT ecosystems, where low latency and efficient resource utilization are critical. However, scheduling aperiodic tasks in fog environments remains a significant challenge, particularly when ensuring strict deadline adherence, energy efficiency, and optimized use of heterogeneous low-power devices. This challenge is further intensified by the dynamic and unpredictable nature of fog networks, making the problem NP-hard. To address these issues, we propose FADDEER (fog adaptive deadline-driven energy efficient reinforcement), a novel deep multi-agent reinforcement learning (Deep MARL)-based scheduling algorithm that integrates earliest deadline first (EDF) and asynchronous advantage actor-critic (A3C) to enable adaptive, scalable, and deadline-aware task scheduling. FADDEER intelligently allocates aperiodic tasks to distributed fog nodes, dynamically balancing task makespan, energy consumption, and resource utilization. Experiments were conducted using the COSCO simulation framework with workloads ranging from 100 to 1,000 aperiodic tasks distributed across fog networks of 10 to 100 nodes, respectively. The results demonstrate that FADDEER consistently outperforms state-of-the-art algorithms, including A3C, multi-agent deep deterministic policy gradient (MADDPG), genetic algorithm, particle swarm optimization (PSO), EDF achieving up to 4% higher deadline adherence and up to 55% lower energy consumption compared to the next best alternatives. Moreover, FADDEER achieved up to 19.4% lower makespan under high task loads and significantly reduced scheduling time, validating its efficiency and responsiveness in dynamic fog environments. These outcomes highlight FADDEER’s robustness, scalability, and effectiveness for real-time IoT task scheduling in fog computing frameworks.</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

Faddeer: a deep multi-agent reinforcement learning-based scheduling algorithm for aperiodic tasks in heterogeneous fog computing networks

  • 期刊论文

  • 发表日期：

    2025年6月16日

查看全文
  • Ganesan Nagabushnam,
  • Kyong Hoon Kim

摘要

Fog computing has become essential for real-time applications in domains such as autonomous driving, smart cities, Industry 4.0, and large-scale IoT ecosystems, where low latency and efficient resource utilization are critical. However, scheduling aperiodic tasks in fog environments remains a significant challenge, particularly when ensuring strict deadline adherence, energy efficiency, and optimized use of heterogeneous low-power devices. This challenge is further intensified by the dynamic and unpredictable nature of fog networks, making the problem NP-hard. To address these issues, we propose FADDEER (fog adaptive deadline-driven energy efficient reinforcement), a novel deep multi-agent reinforcement learning (Deep MARL)-based scheduling algorithm that integrates earliest deadline first (EDF) and asynchronous advantage actor-critic (A3C) to enable adaptive, scalable, and deadline-aware task scheduling. FADDEER intelligently allocates aperiodic tasks to distributed fog nodes, dynamically balancing task makespan, energy consumption, and resource utilization. Experiments were conducted using the COSCO simulation framework with workloads ranging from 100 to 1,000 aperiodic tasks distributed across fog networks of 10 to 100 nodes, respectively. The results demonstrate that FADDEER consistently outperforms state-of-the-art algorithms, including A3C, multi-agent deep deterministic policy gradient (MADDPG), genetic algorithm, particle swarm optimization (PSO), EDF achieving up to 4% higher deadline adherence and up to 55% lower energy consumption compared to the next best alternatives. Moreover, FADDEER achieved up to 19.4% lower makespan under high task loads and significantly reduced scheduling time, validating its efficiency and responsiveness in dynamic fog environments. These outcomes highlight FADDEER’s robustness, scalability, and effectiveness for real-time IoT task scheduling in fog computing frameworks.