Scholarlink

<p>Cloud interconnects are increasingly facing the threat of reliability failures due to traffic surges over shared paths, which cause service-level violations far beyond hardware faults. The August 2025 Cloudflare AWS incident resulted in an operational collapse when a single tenant’s traffic surge caused multiple links to be unavailable for hours, degrading service. This study applies the LSBGO framework as a deterministic scheduler toward this class of reliability failure, through NS-3 simulations of two phases: baseline normal operation and surge conditions. We demonstrate that LSBGO maintains sub-threshold utilization, halves discrepancy indices, reduces packet drop rates by more than two-thirds, and reduces p95 latency violations by over 70% compared to baseline routing schemes. Ablation analysis confirmed that deterministic Latin-square seeding and adaptive genetic refinement are individually insufficient but jointly indispensable, with sensitivity experiments displaying robustness against variations in safety threshold (θ = 0.80–0.95) and slot length (30–120&#xa0;s). These results firmly establish LSBGO as a proactive discrepancy-minimizing control mechanism that could have significantly mitigated the impact of the Cloudflare incident under the modeled conditions.</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

Incident-aware adaptive routing for reliable cloud interconnects: minimizing latency and discrepancy using LSBGO

  • 期刊论文

  • 发表日期：

    2026年4月13日

查看全文
  • Aman Kumar Routh,
  • Prabhat Ranjan

摘要

Cloud interconnects are increasingly facing the threat of reliability failures due to traffic surges over shared paths, which cause service-level violations far beyond hardware faults. The August 2025 Cloudflare AWS incident resulted in an operational collapse when a single tenant’s traffic surge caused multiple links to be unavailable for hours, degrading service. This study applies the LSBGO framework as a deterministic scheduler toward this class of reliability failure, through NS-3 simulations of two phases: baseline normal operation and surge conditions. We demonstrate that LSBGO maintains sub-threshold utilization, halves discrepancy indices, reduces packet drop rates by more than two-thirds, and reduces p95 latency violations by over 70% compared to baseline routing schemes. Ablation analysis confirmed that deterministic Latin-square seeding and adaptive genetic refinement are individually insufficient but jointly indispensable, with sensitivity experiments displaying robustness against variations in safety threshold (θ = 0.80–0.95) and slot length (30–120 s). These results firmly establish LSBGO as a proactive discrepancy-minimizing control mechanism that could have significantly mitigated the impact of the Cloudflare incident under the modeled conditions.