Scholarlink

Containers as a Service (CaaS) allows organisations to manage containers effectively. Containers are lightweight, resource-efficient, portable, and facilitate microservices. In the CaaS model, containers operate on virtual machines (VMs), which are hosted on physical machines. This work seeks to quantify the resources required by a virtual machine to accommodate numerous containers. This work proposes a task grouping methodology based on Reinforcement Learning (RL), taking into account criteria such as task length, submission rate, scheduling class, priority, resource utilisation, latency, and Task Rejection Rate (TRR). VM scaling is an essential activity in which system administrators must optimise resource distribution inside the virtualised environment. We propose employing Deep Convolutional Long Short-Term Memory Networks (Deep-ConvLSTM) for virtual machine sizing. Performance study for VM sizing is performed with 100, 200, and 300 tasks, resulting in best outcomes: resource utilisation of 0.0926 with 300 tasks, TRR of 0.1156 with 100 tasks, and a makespan of 0.5367 with 100 tasks.

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

Reinforcement Learning-Driven Adaptive Task Grouping with Deep-ConvLSTM Based VM Sizing in Container as a Service Cloud Environment

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Kodanda Dhar Naik,
  • Rasmita Panigrahi,
  • Rashmi Ranjan Sahoo

摘要

Containers as a Service (CaaS) allows organisations to manage containers effectively. Containers are lightweight, resource-efficient, portable, and facilitate microservices. In the CaaS model, containers operate on virtual machines (VMs), which are hosted on physical machines. This work seeks to quantify the resources required by a virtual machine to accommodate numerous containers. This work proposes a task grouping methodology based on Reinforcement Learning (RL), taking into account criteria such as task length, submission rate, scheduling class, priority, resource utilisation, latency, and Task Rejection Rate (TRR). VM scaling is an essential activity in which system administrators must optimise resource distribution inside the virtualised environment. We propose employing Deep Convolutional Long Short-Term Memory Networks (Deep-ConvLSTM) for virtual machine sizing. Performance study for VM sizing is performed with 100, 200, and 300 tasks, resulting in best outcomes: resource utilisation of 0.0926 with 300 tasks, TRR of 0.1156 with 100 tasks, and a makespan of 0.5367 with 100 tasks.