Scholarlink

<p>This paper presents a hybrid Machine Learning (ML)-based Virtual Machine Placement (VMP) algorithm for mapping requested VMs into well-suited Physical Machines (PMs) in cloud data centers to meet defined objectives. It considers three prominent conflicting objectives in favor of cloud providers and cloud users simultaneously to make sustainable decisions, which should be aware of data center heterogeneity in terms of processing and memory capacities, and also power consumption usage patterns. Power management and resource dissipation are vital for both green computing objectives and the provider’s cost reduction, whereas utilizing reliable resources is crucial for cloud users, which increases their business continuity. To address the issue, firstly, the power consumption, resource dissipation, and resource reliability models are presented. Then, the VMP issue is formulated into a multi-objective optimization problem with power, dissipation, and reliability optimization viewpoints. The multi-objective Reinforcement Learning-based VMP (MRL-VMP) algorithm is designed to solve the aforementioned combinatorial optimization problem. For each objective, an agent is introduced with a list of possible actions; then, the reaction of the model-free environment is received in the form of a triple (<i>p</i>,<i>d</i>,<i>f</i>), respectively, for power consumption, resource dissipation, and failure probability of resource usage as rewards. In addition, two heuristic procedures are presented to select actions unbiasedly and to measure crowding distance for creating promising solutions in the next episodes. The effectiveness of the proposed MRL-VMP was tested in different scenarios. The simulation results witnessed the superiority of the proposed MRL-VMP against other state-of-the-art methods by about 20.92%, 47.73%, 14.73%, and 12.34% in terms of power management, reliable execution, memory usage, and CPU usage, respectively.</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

Sustainable virtual machine placement in heterogeneous cloud data centers: a reinforcement learning-based approach

  • 期刊论文

  • 发表日期：

    2026年1月3日

查看全文
  • Katayon Dolatshah,
  • Abolfazl Toroghi Haghighat,
  • Vahid Khajehvand,
  • Mirsaeid Hosseini Shirvani

摘要

This paper presents a hybrid Machine Learning (ML)-based Virtual Machine Placement (VMP) algorithm for mapping requested VMs into well-suited Physical Machines (PMs) in cloud data centers to meet defined objectives. It considers three prominent conflicting objectives in favor of cloud providers and cloud users simultaneously to make sustainable decisions, which should be aware of data center heterogeneity in terms of processing and memory capacities, and also power consumption usage patterns. Power management and resource dissipation are vital for both green computing objectives and the provider’s cost reduction, whereas utilizing reliable resources is crucial for cloud users, which increases their business continuity. To address the issue, firstly, the power consumption, resource dissipation, and resource reliability models are presented. Then, the VMP issue is formulated into a multi-objective optimization problem with power, dissipation, and reliability optimization viewpoints. The multi-objective Reinforcement Learning-based VMP (MRL-VMP) algorithm is designed to solve the aforementioned combinatorial optimization problem. For each objective, an agent is introduced with a list of possible actions; then, the reaction of the model-free environment is received in the form of a triple (p,d,f), respectively, for power consumption, resource dissipation, and failure probability of resource usage as rewards. In addition, two heuristic procedures are presented to select actions unbiasedly and to measure crowding distance for creating promising solutions in the next episodes. The effectiveness of the proposed MRL-VMP was tested in different scenarios. The simulation results witnessed the superiority of the proposed MRL-VMP against other state-of-the-art methods by about 20.92%, 47.73%, 14.73%, and 12.34% in terms of power management, reliable execution, memory usage, and CPU usage, respectively.