Scholarlink

<p>Real-time traffic management in smart cities generates massive, latency-sensitive IoT workloads that expose the limitations of both pure cloud and existing fog scheduling approaches. Current methods (e.g., DLSFC, LADFS-QM, QoSPTS) suffer from static data replication, centralized decision-making, oversimplified network models, or lack of explicit DAG dependency handling, resulting in excessive latency, energy waste, and migration overhead under dynamic urban conditions. We propose AMoLAS (<b>A</b>daptive <b>M</b>ulti-<b>O</b>bjective <b>L</b>ocation-<b>A</b>ware <b>S</b>cheduling), a fully decentralized, multi-objective scheduling framework that introduces two novel coupled mechanisms: (i) a Deep Reinforcement Learning (DRL)-based task prioritizer that dynamically ranks interdependent tasks modeled as Directed Acyclic Graphs (DAGs), and (ii) an Improved Jellyfish Search Algorithm (IJFA) that simultaneously optimizes makespan, latency, energy, monetary cost, and migration overhead while adapting replication and placement to real-time access patterns and fluctuating bandwidth/delay. Extensive CloudSim evaluation with 300–1200 tasks and 25–1000&#xa0;MB datasets shows AMoLAS achieves 88% of the optimal ILP bound and outperforms DLSFC, LADFS-QM, and QoSPTS by reducing energy consumption by 18%, end-to-end latency by 22%, operational cost by 20%, and migration cost by 20–35%. These results demonstrate AMoLAS as a scalable and practical solution for latency-critical urban traffic management in fog-cloud-IoT systems.</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 hybrid DRL-IJFA framework for real-time IoT task scheduling in fog-cloud environments

  • 期刊论文

  • 发表日期：

    2026年3月30日

查看全文
  • Yang Li,
  • Iraj Elyasi Komari,
  • Shirin Hasanzadeh

摘要

Real-time traffic management in smart cities generates massive, latency-sensitive IoT workloads that expose the limitations of both pure cloud and existing fog scheduling approaches. Current methods (e.g., DLSFC, LADFS-QM, QoSPTS) suffer from static data replication, centralized decision-making, oversimplified network models, or lack of explicit DAG dependency handling, resulting in excessive latency, energy waste, and migration overhead under dynamic urban conditions. We propose AMoLAS (Adaptive Multi-Objective Location-Aware Scheduling), a fully decentralized, multi-objective scheduling framework that introduces two novel coupled mechanisms: (i) a Deep Reinforcement Learning (DRL)-based task prioritizer that dynamically ranks interdependent tasks modeled as Directed Acyclic Graphs (DAGs), and (ii) an Improved Jellyfish Search Algorithm (IJFA) that simultaneously optimizes makespan, latency, energy, monetary cost, and migration overhead while adapting replication and placement to real-time access patterns and fluctuating bandwidth/delay. Extensive CloudSim evaluation with 300–1200 tasks and 25–1000 MB datasets shows AMoLAS achieves 88% of the optimal ILP bound and outperforms DLSFC, LADFS-QM, and QoSPTS by reducing energy consumption by 18%, end-to-end latency by 22%, operational cost by 20%, and migration cost by 20–35%. These results demonstrate AMoLAS as a scalable and practical solution for latency-critical urban traffic management in fog-cloud-IoT systems.