Scholarlink

<p>Mobile edge computing (MEC) enhances quality of service (QoS) by offloading workloads to nearby edge servers (ESs). The performance of MEC networks critically depends on edge server placement (ESP), the strategic decision of where to locate ESs and how to assign base stations (BSs) to them. While many studies focus on minimizing latency, they often overlook network resilience, creating a critical performance gap. Addressing this, our study presents a framework that formulates ESP as a unified optimization problem to simultaneously minimize access delay, balance workload distribution, and provide fault tolerance. To solve this, we propose an efficient two-step heuristic algorithm. Step one determines the optimal physical locations for ESs by applying the K-medoids clustering algorithm to BS traffic data, placing servers at the center of demand clusters to minimize latency. Step two then implements a resilient assignment mechanism: A custom heuristic allocates each BS to both a primary ES for normal operation and hot-backup ES (dual-active) for failover, thereby embedding fault tolerance directly into the network topology. This integrated approach is governed by constraints on ES capacity, latency thresholds, and service-level agreements (SLAs). Extensive experiments using real-world datasets validate our approach. In a unified benchmark conducted on the Shanghai Telecom dataset, comprising 3000&#xa0;BSs and 300&#xa0;ESs, our Resilient Edge Server Placement (RESP) method achieves the lowest mean access delay of 0.038, with strong workload balance (variance <i>σ</i> = 6.8 × 10<sup>5</sup> requests per minute) and the highest resilience, with a failover success rate (FSR) of 0.90 and a redundancy coverage (RC) of 0.93. This performance approaches the FSR of a robust optimization baseline (0.93) while maintaining a lower delay. Under outage scenarios affecting up to 20% of ESs, RESP sustains an FSR of at least 0.90 for random outages and at least 0.85 for correlated outages, whereas the best competing baseline falls below 0.65. Additionally, RESP’s dual-active operation limits burst queuing delay to below 2&#xa0;s at a load factor (<i>λ</i>) of approximately 0.8. Results demonstrate significant reductions in access delay and workload variance compared to baseline methods. Crucially, our model maintains reliable and responsive performance under simulated ES failure scenarios, confirming the effectiveness of the embedded resilience. These findings underscore the necessity of integrating resilience-aware placement and assignment strategies and highlight RESP’s practical impact for mission-critical, latency-sensitive MEC deployments.</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

Resilient and efficient edge server placement in mobile edge computing: a dual-strategy approach for optimized performance and fault tolerance

  • 期刊论文

  • 发表日期：

    2025年9月9日

查看全文
  • Arman Sanaei Asl,
  • Massoud Reza Hashemi

摘要

Mobile edge computing (MEC) enhances quality of service (QoS) by offloading workloads to nearby edge servers (ESs). The performance of MEC networks critically depends on edge server placement (ESP), the strategic decision of where to locate ESs and how to assign base stations (BSs) to them. While many studies focus on minimizing latency, they often overlook network resilience, creating a critical performance gap. Addressing this, our study presents a framework that formulates ESP as a unified optimization problem to simultaneously minimize access delay, balance workload distribution, and provide fault tolerance. To solve this, we propose an efficient two-step heuristic algorithm. Step one determines the optimal physical locations for ESs by applying the K-medoids clustering algorithm to BS traffic data, placing servers at the center of demand clusters to minimize latency. Step two then implements a resilient assignment mechanism: A custom heuristic allocates each BS to both a primary ES for normal operation and hot-backup ES (dual-active) for failover, thereby embedding fault tolerance directly into the network topology. This integrated approach is governed by constraints on ES capacity, latency thresholds, and service-level agreements (SLAs). Extensive experiments using real-world datasets validate our approach. In a unified benchmark conducted on the Shanghai Telecom dataset, comprising 3000 BSs and 300 ESs, our Resilient Edge Server Placement (RESP) method achieves the lowest mean access delay of 0.038, with strong workload balance (variance σ = 6.8 × 105 requests per minute) and the highest resilience, with a failover success rate (FSR) of 0.90 and a redundancy coverage (RC) of 0.93. This performance approaches the FSR of a robust optimization baseline (0.93) while maintaining a lower delay. Under outage scenarios affecting up to 20% of ESs, RESP sustains an FSR of at least 0.90 for random outages and at least 0.85 for correlated outages, whereas the best competing baseline falls below 0.65. Additionally, RESP’s dual-active operation limits burst queuing delay to below 2 s at a load factor (λ) of approximately 0.8. Results demonstrate significant reductions in access delay and workload variance compared to baseline methods. Crucially, our model maintains reliable and responsive performance under simulated ES failure scenarios, confirming the effectiveness of the embedded resilience. These findings underscore the necessity of integrating resilience-aware placement and assignment strategies and highlight RESP’s practical impact for mission-critical, latency-sensitive MEC deployments.