CDN Server Selection Policies

💡 The Core Idea

When a user requests content, the CDN must select a specific Cluster (location) and a specific Server (machine).

🧠 Cluster Selection (Which Location?)

1. Geo-Proximity: Pick the geographically closest cluster “as the crow flies.”
   • Flaw: Closest geo-location might have a congested network path or longer BGP route.
   • Flaw: Decisions are often based on the user’s LDNS location, not the user’s actual IP.
2. Real-time Measurements:
   • Active: LDNS pings multiple clusters (rare/traffic heavy).
   • Passive: The CDN monitors TCP connection stats from existing users in the same subnet to predict performance.

🧠 Server Selection (Which Machine?)

Once inside the cluster, simple load balancing isn’t enough because not all servers hold all content (disk space limits).

• Content-Aware Hashing: Requests for specific content are hashed to specific servers (via Consistent Hashing) to maximize cache hit rates.

🔗 Connections

• Source: Source - CDNs and Overlay Networks