Chapter 1: What Edge Computing Is and Where the Edge Lives
The edge is not a place; it is a latency budget. For interactive applications, the round-trip budget between a user gesture and a visible response is 100 milliseconds — anything longer and the user perceives the system as sluggish. The traditional centralized cloud model, with round trips of 50 to 200 milliseconds to the nearest region, leaves almost no headroom inside that budget once you account for the application logic itself.
Where the edge actually sits
Three practical tiers exist for edge infrastructure:
1. CDN edge nodes — Cloudflare, Fastly, Akamai. ~50ms from most users. Best for HTTP request/response, JWT verification, A/B routing, and stateless transforms. 2. Regional points of presence — AWS Local Zones, Azure Edge Zones. ~20-30ms from metropolitan areas. Best for workloads that need a real database read or moderate compute. 3. Device edge — phones, cars, edge gateways. 0ms from the user but limited compute and intermittent connectivity.
Latency math
A user in Berlin tapping a button on a mobile app will see:
- Mobile network RTT: 20-40ms
- CDN edge processing: 5-15ms
- Database roundtrip to a centralized region: 60-120ms
- Total: 85-175ms
Move the database read to a regional PoP with a read replica, and the same flow drops to 25-75ms total — well inside the 100ms budget.
The edge is not free
Edge compute costs roughly 3-5x per CPU-second compared to centralized regions. The economics work when the latency win translates to either higher conversion (retail, gaming) or lower bandwidth (video, IoT). They do not work for batch processing or anything that does not need to respond to a user in real time.
The decision framework
Pick the edge tier that matches your latency budget, not the tier that is "closest to the user". For most consumer applications, CDN edge is the right starting point. For industrial control or autonomous systems, device edge is the only answer. Regional PoPs are the middle ground that few teams actually need.