Scholarlink

<p>The consolidation of Virtualized Network Functions (VNFs) in Virtual Machines (VMs) requires careful consideration of predefined VNF ordering to avoid violating Service Level Agreements (SLAs) and ensure proper Service Function Chain (SFC) generation. Existing approaches often fail to address dynamic network conditions and multi-objective optimization. In this paper, we propose a novel hybrid approach integrating <i>Fuzzy Formal Concept Analysis (Fuzzy-FCA)</i> with a <i>Fuzzy Inference System (FIS)</i> for VNF placement constraints, and a <i>Parallel Bi-state Deep Reinforcement Learning (PBDRL)</i> algorithm for VNF chaining and migration. Our comprehensive evaluation demonstrates that the proposed approach achieves <i>23.4% lower reallocation cost</i> compared to Integer Linear Programming (ILP)-based approaches, <i>25.1% improvement</i> over standalone PBDRL approaches, <i>40–50% faster computation times</i> across varying network scales <i>18.7% reduction in end-to-end latency</i> while maintaining QoS requirements and Balanced resource utilization with <i>92.3% average VM utilization rate</i>. Experimental results across multiple topologies (Barabasi-Albert (BA), Waxman (WA) and Erdos-Renyi (ER)) validate the robustness and efficiency of our hybrid framework in dynamic network 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

Deep Learning and Fuzzy Logic Integration for Virtual Network Function Deployment

  • 期刊论文

  • 发表日期：

    2026年2月18日

查看全文
  • Jalel Eddine Hajlaoui,
  • Wided Khemili,
  • Mohamed Nazih Omri

摘要

The consolidation of Virtualized Network Functions (VNFs) in Virtual Machines (VMs) requires careful consideration of predefined VNF ordering to avoid violating Service Level Agreements (SLAs) and ensure proper Service Function Chain (SFC) generation. Existing approaches often fail to address dynamic network conditions and multi-objective optimization. In this paper, we propose a novel hybrid approach integrating Fuzzy Formal Concept Analysis (Fuzzy-FCA) with a Fuzzy Inference System (FIS) for VNF placement constraints, and a Parallel Bi-state Deep Reinforcement Learning (PBDRL) algorithm for VNF chaining and migration. Our comprehensive evaluation demonstrates that the proposed approach achieves 23.4% lower reallocation cost compared to Integer Linear Programming (ILP)-based approaches, 25.1% improvement over standalone PBDRL approaches, 40–50% faster computation times across varying network scales 18.7% reduction in end-to-end latency while maintaining QoS requirements and Balanced resource utilization with 92.3% average VM utilization rate. Experimental results across multiple topologies (Barabasi-Albert (BA), Waxman (WA) and Erdos-Renyi (ER)) validate the robustness and efficiency of our hybrid framework in dynamic network environments.