Downtime Management in Multi-CDN Steering Systems

Managing CDN failures in multi-CDN streaming environments requires a careful balance between maintaining Quality of Experience (QoE) and ensuring timely detection of CDN recovery. Traditional methods rely on external probes or client-based monitoring, both of which have limitations in accuracy or feasibility. This paper explores an alternative approach that strategically assigns a small subset of users to a failing CDN while dynamically adjusting their Time-To-Live (TTL) based on buffer length information. By leveraging buffer-aware content steering, the system ensures that only users with sufficient buffer capacity are temporarily assigned to the faulty CDN, minimizing rebuffering events while maintaining continuous monitoring of the CDN’s recovery status.

Through simulations, we evaluate the trade-offs between different user assignment strategies and demonstrate that a Rotating Sacrifice with Buffer-Based TTL approach provides an effective balance. This method achieves rapid recovery detection, typically within seconds, while keeping QoE degradation at levels comparable to a baseline approach where no monitoring is performed. The findings highlight the benefits of incorporating buffer length data into the content steering process, leading to a fairer and more efficient multi-CDN orchestration. We advocate for the inclusion of buffer-level reporting as a standard parameter in content steering systems to improve resilience and service continuity in large-scale streaming operations.

Gwendal Simon | Synamedia | Rennes, France