Scholarlink

<p>High demand for cloud services and resources has led to increased energy consumption in cloud data centers. Various approaches have been proposed that employ heuristic algorithms to consolidate virtual machines (VMs) into a minimum number of physical machines (PMs). While some VMs are CPU-intensive and others are memory-intensive, there is currently no algorithm that considers VM consolidation based on their types. Therefore, we propose a novel scheduling algorithm that applies machine learning techniques to classify VMs according to their resource demands. This classification is achieved by observing their resource usage for a short period after their arrival. The algorithm then predicts the future required resources for VMs and redeploys them to a PM with a mix of types of VMs. Our proposed algorithm maximizes the utilization of the resources of PMs to reduce the number of active PMs and minimize service level agreement violations (SLAv). Our proposed algorithm was tested on a dataset that was extracted from Google’s workload Trace. Our algorithm achieved a mean 40% reduction in energy consumption and a 12% increase in resource utilization compared to the benchmark algorithms used in this study. In addition, and comparing to the Google scheduler, our algorithm reduced energy consumption by 51% and increased resource utilization by 62.43%.</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

A machine learning technique for optimizing virtual machine placement in data-centres

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月7日

查看全文
  • Abdullah Alelyani,
  • Amitava Datta,
  • Ghulam Mubashar Hassan

摘要

High demand for cloud services and resources has led to increased energy consumption in cloud data centers. Various approaches have been proposed that employ heuristic algorithms to consolidate virtual machines (VMs) into a minimum number of physical machines (PMs). While some VMs are CPU-intensive and others are memory-intensive, there is currently no algorithm that considers VM consolidation based on their types. Therefore, we propose a novel scheduling algorithm that applies machine learning techniques to classify VMs according to their resource demands. This classification is achieved by observing their resource usage for a short period after their arrival. The algorithm then predicts the future required resources for VMs and redeploys them to a PM with a mix of types of VMs. Our proposed algorithm maximizes the utilization of the resources of PMs to reduce the number of active PMs and minimize service level agreement violations (SLAv). Our proposed algorithm was tested on a dataset that was extracted from Google’s workload Trace. Our algorithm achieved a mean 40% reduction in energy consumption and a 12% increase in resource utilization compared to the benchmark algorithms used in this study. In addition, and comparing to the Google scheduler, our algorithm reduced energy consumption by 51% and increased resource utilization by 62.43%.