Scholarlink

<p>Cloud computing provides high-performance computing and storage services, but efficient task scheduling between virtual machines (VMs) remains challenging because of its changing characteristics. This paper proposes a fused attention rectified linear unit enhanced actor–critic graph reinforcement learning method that optimizes task execution on VMs, minimizing makespan and cost. This method leverages advanced technologies, including graph neural networks (GNNs), attention mechanisms and deep reinforcement learning (DRL). First, the GNN effectively processes complex network structures. A heterogeneous graph embedding framework integrates task requirements, VM configurations, resource dependencies, and “VM–Task” constraints to model dynamic cloud resource states. Second, the attention mechanism efficiently extracts graph node features. A GNN with a fused attention rectified linear unit extracts feature vectors from VMs and task nodes. This mechanism mitigates gradient vanishing during feature extraction while emphasizing critical “VM–Task” mapping relationships. Third, DRL enables rapid adaptability and real-time decision-making. An actor–critic reinforcement learning framework employs these feature vectors as states and actions to develop adaptive evaluation metrics and reward functions that balance competing optimization objectives. Through iterative policy updates, the agent learns optimal “VM–Task” deployment strategies. Finally, experiments results show that compared with state-of-the-art algorithms under different conditions, the proposed method reduces makespans by approximately 10.94%, cost by approximately 9.29%.</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

Fused attention rectified linear unit–empowered graph reinforcement learning for task scheduling in the cloud

  • 期刊论文

  • 发表日期：

    2025年9月19日

查看全文
  • Xiaoxian Yang,
  • Zhifeng Wang,
  • Jiacheng Li,
  • Lin Chen

摘要

Cloud computing provides high-performance computing and storage services, but efficient task scheduling between virtual machines (VMs) remains challenging because of its changing characteristics. This paper proposes a fused attention rectified linear unit enhanced actor–critic graph reinforcement learning method that optimizes task execution on VMs, minimizing makespan and cost. This method leverages advanced technologies, including graph neural networks (GNNs), attention mechanisms and deep reinforcement learning (DRL). First, the GNN effectively processes complex network structures. A heterogeneous graph embedding framework integrates task requirements, VM configurations, resource dependencies, and “VM–Task” constraints to model dynamic cloud resource states. Second, the attention mechanism efficiently extracts graph node features. A GNN with a fused attention rectified linear unit extracts feature vectors from VMs and task nodes. This mechanism mitigates gradient vanishing during feature extraction while emphasizing critical “VM–Task” mapping relationships. Third, DRL enables rapid adaptability and real-time decision-making. An actor–critic reinforcement learning framework employs these feature vectors as states and actions to develop adaptive evaluation metrics and reward functions that balance competing optimization objectives. Through iterative policy updates, the agent learns optimal “VM–Task” deployment strategies. Finally, experiments results show that compared with state-of-the-art algorithms under different conditions, the proposed method reduces makespans by approximately 10.94%, cost by approximately 9.29%.