Scholarlink

<p>Energy-aware Virtual Machine Placement (VMP) represents an important optimization task in cloud data centers. Energy waste and poor utilization of resources in such systems occur due to inappropriate allocation of VMs to PMs. Available heuristics and evolutionary methods may face problems of premature convergence and low diversity of search process, especially in cases of solving large scale heterogeneous problems. This paper suggests Loser-Augmented Duelist Algorithm (LADA), which is a modified evolutionary meta-heuristic extending classical Duelist Algorithm with loser learning strategies, champion-based tuning, historical memory keeping, diversity-driven adaptive mutation, and elite local search mechanisms. This algorithm intends to maintain the balance between exploration and exploitation processes as well as solution diversity during optimization procedure. Experiments were carried out on eighteen synthetic benchmarks as well as on Google Cluster workloads with different restrictions of CPU, memory, and bandwidth capacity. Five different random seeds were used as starting points for each stochastic algorithm. Results show that LADA outperforms well-established approaches, including First Fit Decreasing (FFD), Simulated Annealing (SA), Particle Swarm Optimization (PSO), Differential Evolution (DE), Duelist Algorithm (DA), Hybrid Grey Wolf Optimizer (HGWO), and Hybrid Ant Colony—Particle Swarm Optimization (HAPSO) in terms of energy consumption. In some of medium and large benchmark instances, LADA demonstrates less energy usage compared to competitive algorithms along with faster convergence rate and better solution stability. Statistical analysis of results based on Friedman tests, Holm-Bonferroni adjusted pairwise comparison, and effect size estimation proves the significance of the observed difference. Moreover, ablation, sensitivity, and scalability study confirms the effectiveness and efficiency of the proposed approach.</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 loser-augmented duelist algorithm for cloud data center energy consumption minimization

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月29日

查看全文
  • Amol C. Adamuthe,
  • Vijay H. Kalmani,
  • Pooja Bagane

摘要

Energy-aware Virtual Machine Placement (VMP) represents an important optimization task in cloud data centers. Energy waste and poor utilization of resources in such systems occur due to inappropriate allocation of VMs to PMs. Available heuristics and evolutionary methods may face problems of premature convergence and low diversity of search process, especially in cases of solving large scale heterogeneous problems. This paper suggests Loser-Augmented Duelist Algorithm (LADA), which is a modified evolutionary meta-heuristic extending classical Duelist Algorithm with loser learning strategies, champion-based tuning, historical memory keeping, diversity-driven adaptive mutation, and elite local search mechanisms. This algorithm intends to maintain the balance between exploration and exploitation processes as well as solution diversity during optimization procedure. Experiments were carried out on eighteen synthetic benchmarks as well as on Google Cluster workloads with different restrictions of CPU, memory, and bandwidth capacity. Five different random seeds were used as starting points for each stochastic algorithm. Results show that LADA outperforms well-established approaches, including First Fit Decreasing (FFD), Simulated Annealing (SA), Particle Swarm Optimization (PSO), Differential Evolution (DE), Duelist Algorithm (DA), Hybrid Grey Wolf Optimizer (HGWO), and Hybrid Ant Colony—Particle Swarm Optimization (HAPSO) in terms of energy consumption. In some of medium and large benchmark instances, LADA demonstrates less energy usage compared to competitive algorithms along with faster convergence rate and better solution stability. Statistical analysis of results based on Friedman tests, Holm-Bonferroni adjusted pairwise comparison, and effect size estimation proves the significance of the observed difference. Moreover, ablation, sensitivity, and scalability study confirms the effectiveness and efficiency of the proposed approach.