Scholarlink

<p>Mission-critical services increasingly rely on computing continuum that demand ultra-low latency and continuous availability. Existing methods often face delays in decision-making and slow adaptation to failures, particularly across dynamic cloud-to-edge environments including IoT layers. Addressing these challenges, we propose a novel service placement strategy using deep reinforcement learning with Proximal Policy Optimization (PPO), enhanced by transfer learning for faster adaptation in successive placement stages. A lightweight fault-tolerance mechanism further ensures reliable, uninterrupted service delivery. Simulation results show that our method outperforms state-of-the-art approaches in meeting deadline and reliability requirements, while effective load distribution improves on-time responses. The main contribution is a unified framework that combines intelligent, adaptive placement with fault tolerance, enabling dependable and timely provisioning of mission-critical services across the computing continuum.</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 placement method based on deep reinforcement learning for mission critical services in computing continuum

  • 期刊论文

  • 发表日期：

    2026年5月29日

查看全文
  • Reza Sookhtsaraei,
  • Mehdi Sakhaei-nia,
  • Fereshteh-Azadi Parand

摘要

Mission-critical services increasingly rely on computing continuum that demand ultra-low latency and continuous availability. Existing methods often face delays in decision-making and slow adaptation to failures, particularly across dynamic cloud-to-edge environments including IoT layers. Addressing these challenges, we propose a novel service placement strategy using deep reinforcement learning with Proximal Policy Optimization (PPO), enhanced by transfer learning for faster adaptation in successive placement stages. A lightweight fault-tolerance mechanism further ensures reliable, uninterrupted service delivery. Simulation results show that our method outperforms state-of-the-art approaches in meeting deadline and reliability requirements, while effective load distribution improves on-time responses. The main contribution is a unified framework that combines intelligent, adaptive placement with fault tolerance, enabling dependable and timely provisioning of mission-critical services across the computing continuum.