Scholarlink

<p>In cloud computing and smart manufacturing, effective workflow scheduling is vital for balancing execution time, energy consumption, and system responsiveness. This gap motivates the development of intelligent, adaptive approaches that can jointly optimize conflicting objectives under uncertainty. To address these, we propose VECO-GNN—a novel, end-to-end workflow scheduling framework that integrates Variational Auto-encoding (VAE), Collaborative Multi-Objective Optimization (CMOO), and Graph Neural Network (GNN) to perform mapping of virtual machines (VMs). The VAE captures compact and uncertainty-aware representations of task features, while CMOO explores Pareto-optimal mappings across three objectives: execution time, energy consumption, and decision latency. The GNN models task–VM interactions through edge embeddings, enabling relational reasoning for high-quality scheduling decisions. Our methodology is evaluated across five datasets: GENOME, LIGO, AUDIO, RNA, and a synthetic benchmark (Synthetic-DS) designed using probabilistic Compared to DRL-GNN, Random, and heuristic baselines, VECO-GNN consistently outperforms in key metrics. Moreover, VECO-GNN records the highest hyper volume (0.631), lowest Inverted generational distance (IGD 0.078), and widest Pareto coverage (72.4%), reflecting its strong multi-objective optimization capabilities. This empirical evidence demonstrates VECO-GNN’s robustness, scalability, and adaptability in diverse, high-dimensional scheduling 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

VECO-GNN: variational encoding and graph neural optimization framework for multi-objective workflow scheduling in cloud environment

  • 期刊论文

  • 发表日期：

    2025年11月21日

查看全文
  • Vivek Kumar,
  • Ram Krishan

摘要

In cloud computing and smart manufacturing, effective workflow scheduling is vital for balancing execution time, energy consumption, and system responsiveness. This gap motivates the development of intelligent, adaptive approaches that can jointly optimize conflicting objectives under uncertainty. To address these, we propose VECO-GNN—a novel, end-to-end workflow scheduling framework that integrates Variational Auto-encoding (VAE), Collaborative Multi-Objective Optimization (CMOO), and Graph Neural Network (GNN) to perform mapping of virtual machines (VMs). The VAE captures compact and uncertainty-aware representations of task features, while CMOO explores Pareto-optimal mappings across three objectives: execution time, energy consumption, and decision latency. The GNN models task–VM interactions through edge embeddings, enabling relational reasoning for high-quality scheduling decisions. Our methodology is evaluated across five datasets: GENOME, LIGO, AUDIO, RNA, and a synthetic benchmark (Synthetic-DS) designed using probabilistic Compared to DRL-GNN, Random, and heuristic baselines, VECO-GNN consistently outperforms in key metrics. Moreover, VECO-GNN records the highest hyper volume (0.631), lowest Inverted generational distance (IGD 0.078), and widest Pareto coverage (72.4%), reflecting its strong multi-objective optimization capabilities. This empirical evidence demonstrates VECO-GNN’s robustness, scalability, and adaptability in diverse, high-dimensional scheduling environments.