CDN Architecture

Overview

Ether Networks operated a respectable single-region origin and a growing customer base that increasingly lived outside that region. Latency complaints were the symptom; the actual condition was an architecture that had never been asked to think about geography. GRAY was brought in to design the edge network from first principles, not to bolt a CDN vendor onto the existing stack.

We modelled traffic against fourteen candidate points of presence and selected a topology that balanced population density, existing carrier peering, and cross-border data regulation — not simply 'closest node wins'. Each region runs its own cache tier with independent invalidation, so a purge in one market never triggers unnecessary churn in another.

Failover was designed as a first-class behaviour rather than a fallback afterthought: health checks run at the edge, not just at origin, and traffic reroutes at the DNS and anycast layer within single-digit seconds of a regional fault — verified under live chaos testing before the system was allowed anywhere near production traffic.

The engineering work was paired with a rebuilt marketing surface, because a 14-region network is only a story if customers can actually see it. The same design tokens, grid, and typographic system that shaped the internal network diagrams shaped the public site — infrastructure and its explanation sharing one visual language rather than two disconnected artifacts from two disconnected teams.

The network now delivers to fourteen regions at sub-50ms median latency, with the same team maintaining both the platform and the site that describes it. Ether Networks did not just get faster delivery — they got a single coherent story about their infrastructure, told consistently from the terminal to the homepage.