Scholarlink

<p>The rapid expansion of IoT applications has led to massive, highly dynamic workloads that frequently overload fog nodes, demanding real-time and computationally intensive load balancing strategies. To address these challenges, scalable, parallel, and high-performance computing (HPC) techniques are required which are capable of handling large-volume data and time-critical decision-making. This study introduces an application-driven container workload dataset generation using virtual machines, incorporating parameters such as CPU usage, RAM, number of cores, and memory allocation to support large-scale host workload analysis. In order to categorize hosts as overburdened, underloaded, or balanced, an ensemble clustering approach is employed with various preprocessing techniques. The proposed ensemble clustering method achieves the highest silhouette coefficient (0.964), surpassing individual clustering algorithms. The selected ensemble approach effectively transforms unlabeled data into labeled datasets, enabling efficient workload migration from overloaded to underloaded hosts. Containers are used to efficiently distribute workload as a result of their lightweight nature. The most appropriate containers for migration are identified using a computationally intensive container selection algorithm. Further, this study introduces a novel Hybrid Beluga Whale Optimization algorithm with Simulated Annealing (HBWO-SA), a metaheuristic approach used to balance load and reduce container migration cost. This proposed HBWO-SA algorithm is designed for parallelizable execution and benefits from HPC platforms in accelerating convergence and improving optimization efficiency. Experimental evaluations demonstrate that the proposed HBWO-SA algorithm significantly reduces latency, enhances resource efficiency, and outperforms existing metaheuristic techniques, including ABC, BWO, E-ABC, Firefly, GWO, and DIWPSO across multiple evaluation metrics. Overall, the proposed approach provides an HPC-aligned, scalable solution for real-time resource management and performance optimization in fog environment. It significantly reduces the ratio of load imbalance and rate of container migration compared to existing techniques.</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

Workload consolidation in fog computing: an ensemble clustering and hybrid beluga whale-simulated annealing optimization approach

  • 期刊论文

  • 发表日期：

    2026年1月16日

查看全文
  • Shabnam Bawa,
  • RajKumar Tekchandani,
  • Prashant Singh Rana

摘要

The rapid expansion of IoT applications has led to massive, highly dynamic workloads that frequently overload fog nodes, demanding real-time and computationally intensive load balancing strategies. To address these challenges, scalable, parallel, and high-performance computing (HPC) techniques are required which are capable of handling large-volume data and time-critical decision-making. This study introduces an application-driven container workload dataset generation using virtual machines, incorporating parameters such as CPU usage, RAM, number of cores, and memory allocation to support large-scale host workload analysis. In order to categorize hosts as overburdened, underloaded, or balanced, an ensemble clustering approach is employed with various preprocessing techniques. The proposed ensemble clustering method achieves the highest silhouette coefficient (0.964), surpassing individual clustering algorithms. The selected ensemble approach effectively transforms unlabeled data into labeled datasets, enabling efficient workload migration from overloaded to underloaded hosts. Containers are used to efficiently distribute workload as a result of their lightweight nature. The most appropriate containers for migration are identified using a computationally intensive container selection algorithm. Further, this study introduces a novel Hybrid Beluga Whale Optimization algorithm with Simulated Annealing (HBWO-SA), a metaheuristic approach used to balance load and reduce container migration cost. This proposed HBWO-SA algorithm is designed for parallelizable execution and benefits from HPC platforms in accelerating convergence and improving optimization efficiency. Experimental evaluations demonstrate that the proposed HBWO-SA algorithm significantly reduces latency, enhances resource efficiency, and outperforms existing metaheuristic techniques, including ABC, BWO, E-ABC, Firefly, GWO, and DIWPSO across multiple evaluation metrics. Overall, the proposed approach provides an HPC-aligned, scalable solution for real-time resource management and performance optimization in fog environment. It significantly reduces the ratio of load imbalance and rate of container migration compared to existing techniques.