Scholarlink

<p>Workflow scheduling in Infrastructure-as-a-Service (IaaS) cloud environments presents significant challenges due to the heterogeneity and dynamic variability of virtualized resources. Traditional approaches often overlook the detailed relationship between specific task requirements and virtual machine (VM) configurations. In this study, we define these relationships as sensitivity. The paper introduces a sensitivity-aware scheduling framework that optimally pairs tasks with VMs based on their resource sensitivities. A Deep Q-Network (DQN)-based reinforcement learning model is proposed to dynamically allocate tasks and adapt VM resource configurations, thereby minimizing both execution time and resource over-provisioning. Experimental results demonstrate that the proposed OPDQN (Optimally Pairs and Deep Q-Network) method consistently outperforms state-of-the-art algorithms such as NHGCPM, VPSS, and DARS in terms of average execution time, CPU resource consumption and alignment between task sensitivities and virtual machine configurations across varying workflow complexities.</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

Workflow scheduling in IaaS clouds with the optimal pairing between tasks and virtual machines

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年9月24日

查看全文
  • Xianmei Hua,
  • Lubin Zheng

摘要

Workflow scheduling in Infrastructure-as-a-Service (IaaS) cloud environments presents significant challenges due to the heterogeneity and dynamic variability of virtualized resources. Traditional approaches often overlook the detailed relationship between specific task requirements and virtual machine (VM) configurations. In this study, we define these relationships as sensitivity. The paper introduces a sensitivity-aware scheduling framework that optimally pairs tasks with VMs based on their resource sensitivities. A Deep Q-Network (DQN)-based reinforcement learning model is proposed to dynamically allocate tasks and adapt VM resource configurations, thereby minimizing both execution time and resource over-provisioning. Experimental results demonstrate that the proposed OPDQN (Optimally Pairs and Deep Q-Network) method consistently outperforms state-of-the-art algorithms such as NHGCPM, VPSS, and DARS in terms of average execution time, CPU resource consumption and alignment between task sensitivities and virtual machine configurations across varying workflow complexities.