Scholarlink

<p>Edge computing based on Software Defined Network (SDN) offers the potential for efficient utilization of limited computing and network resources. Network Function Virtualization (NFV) enables network resource allocation on demand and service functions to run on commodity servers. A user request usually consists of multiple Virtual Network Functions (VNFs) in a specific order and can be considered as a Service Function Chain (SFC). As GPU can accelerate the processing of certain VNFs, choosing the appropriate user requests to accelerate is very important for the system performance. For randomly arrived requests, this paper presents weak- and strong-affinity-aware request allocation approaches to optimize energy efficiency, by formulating user request allocation problems as integer programming models. Furthermore, this paper proposes a heuristic weak-affinity-aware request allocation algorithm (Greedy-weak-affinity), which combines randomized rounding and greedy strategies to significantly reduce computational complexity. The theoretical performance bounds of Greedy-weak-affinity are rigorously proven. Simulations show that under high-load conditions, the Greedy-weak-affinity algorithm achieves approximately 27% higher energy efficiency and 25% higher throughput, compared to the benchmark prioritizing maximum-capacity edge nodes.</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

Affinity-aware dynamic request allocation for SDN-based edge computing

  • 期刊论文

  • 发表日期：

    2025年9月29日

查看全文
  • Shoulu Hou,
  • Min Gan,
  • Yaru Zhao,
  • Kailan Zhao,
  • Qiang Tong,
  • Xiulei Liu

摘要

Edge computing based on Software Defined Network (SDN) offers the potential for efficient utilization of limited computing and network resources. Network Function Virtualization (NFV) enables network resource allocation on demand and service functions to run on commodity servers. A user request usually consists of multiple Virtual Network Functions (VNFs) in a specific order and can be considered as a Service Function Chain (SFC). As GPU can accelerate the processing of certain VNFs, choosing the appropriate user requests to accelerate is very important for the system performance. For randomly arrived requests, this paper presents weak- and strong-affinity-aware request allocation approaches to optimize energy efficiency, by formulating user request allocation problems as integer programming models. Furthermore, this paper proposes a heuristic weak-affinity-aware request allocation algorithm (Greedy-weak-affinity), which combines randomized rounding and greedy strategies to significantly reduce computational complexity. The theoretical performance bounds of Greedy-weak-affinity are rigorously proven. Simulations show that under high-load conditions, the Greedy-weak-affinity algorithm achieves approximately 27% higher energy efficiency and 25% higher throughput, compared to the benchmark prioritizing maximum-capacity edge nodes.