Scholarlink

<p>The growth of latency-sensitive Internet of Things (IoT) applications has intensified the need for efficient task offloading in edge computing environments. Existing metaheuristic-based solutions often suffer from poor initialization and premature convergence, resulting in inefficient resource utilization. This paper presents a two-stage task offloading framework that integrates the Best Edge VM Shortest Task (BEVST) initialization strategy with the Lévy-Seagull Optimization Algorithm (Lévy-SOA). BEVST provides a structure-aware warm start by assigning short tasks to suitable edge virtual machines, enabling balanced workload distribution. Lévy-SOA then refines task execution using Lévy flight-based exploration to escape local optima. Experimental results show that the proposed framework reduces makespan by about 13%, task rejection ratio by 25%, execution cost by 7%, and energy consumption by 15% compared to conventional methods, demonstrating its effectiveness for dynamic IoT–edge–cloud environments.</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 two-stage task offloading framework for IoT edge computing using BEVST initialization and Lévy-SOA optimization

  • 期刊论文

  • 发表日期：

    2026年5月13日

查看全文
  • Avishek Sinha,
  • Samayveer Singh

摘要

The growth of latency-sensitive Internet of Things (IoT) applications has intensified the need for efficient task offloading in edge computing environments. Existing metaheuristic-based solutions often suffer from poor initialization and premature convergence, resulting in inefficient resource utilization. This paper presents a two-stage task offloading framework that integrates the Best Edge VM Shortest Task (BEVST) initialization strategy with the Lévy-Seagull Optimization Algorithm (Lévy-SOA). BEVST provides a structure-aware warm start by assigning short tasks to suitable edge virtual machines, enabling balanced workload distribution. Lévy-SOA then refines task execution using Lévy flight-based exploration to escape local optima. Experimental results show that the proposed framework reduces makespan by about 13%, task rejection ratio by 25%, execution cost by 7%, and energy consumption by 15% compared to conventional methods, demonstrating its effectiveness for dynamic IoT–edge–cloud environments.