Edge Computing Architecture Cheat Sheet

Edge computing architecture and patterns — CDNs, edge functions, IoT gateways, fog computing, data locality, and the trade-offs between edge and cloud processing.

Last Updated: May 1, 2025

Edge Computing Layers

Layer	Location	Latency	Examples	Best For
Device Edge	On the device itself	< 1ms	Sensor processors, phone ML, microcontrollers	Real-time control, offline-first, privacy-sensitive
On-Premise Edge	Factory floor, retail store	1-5ms	AWS Outposts, Azure Stack, local K8s	Industrial IoT, retail POS, hospital systems
Network Edge	Cell tower, cable headend	5-20ms	AWS Wavelength, Azure Edge Zones	5G applications, AR/VR, autonomous vehicles
Regional Edge	Metro data centers	10-50ms	Cloudflare, Fastly, Akamai CDNs	CDN, edge functions, A/B testing at edge
Cloud Core	Traditional cloud regions	50-200ms	AWS us-east-1, GCP us-central1	Heavy compute, big data, ML training, archival

CDN & Edge Functions

Cloudflare Workers

JavaScript/TypeScript at 300+ PoPs. ~0ms cold start. WebSocket, KV, Durable Objects. Deploy globally in seconds.

AWS Lambda@Edge

Node.js/Python at CloudFront edge locations. Trigger: Viewer Request, Origin Request, Origin Response, Viewer Response. Max 30s execution.

Vercel Edge Functions

Next.js edge runtime. Deno-based. Runs at Vercel's edge network. Great for SSR, API routes, A/B testing, geo-routing.

Fastly Compute@Edge

WebAssembly-based (Rust, JS, Go). Sub-millisecond cold starts. Full request lifecycle control. Advanced caching with VCL or custom logic.

Deno Deploy

Edge JavaScript/TypeScript. Built on Deno. Global by default. Web standard APIs. KV store, queues, cron jobs. Built-in observability.

Edge Patterns & Use Cases

Item	Description
`Geo-Routing`	Route user to nearest backend region based on IP. Edge function reads CF-IPCountry header, redirects to region-specific origin. Reduces RTT by 30-70%.
`Edge Authentication`	Validate JWT at edge before request hits origin. Reject invalid tokens at CDN layer. Reduces origin load and improves attacker DDoS resilience.
`A/B Testing at Edge`	Cookie-based traffic splitting: 50% to origin-v1, 50% to origin-v2. No client-side flicker. Experiment runs without app code changes.
`Edge-Side Includes (ESI)`	Compose pages from cached fragments at CDN. Static layout + dynamic user data. Varnish/Cloudflare: assemble personalized pages from shared cached components.
`IoT Data Preprocessing`	Filter and aggregate sensor data at the edge before sending to cloud. Reduce bandwidth costs by 90%+: only send anomalies or 5-min rollups, not raw streams.
`Edge ML Inference`	Run TensorFlow Lite, ONNX Runtime, or edge-optimized models. Privacy-preserving (data never leaves device). Applications: voice assistants, visual inspection, predictive maintenance.

Edge vs Cloud Trade-offs

Item	Description
`Latency`	Edge: 1-50ms. Cloud: 50-200ms+. For real-time like autonomous driving or VR, edge is non-negotiable.
`Compute Power`	Edge: constrained CPU/RAM (10ms-30s timeouts). Cloud: unlimited scale. Heavy ML training or big data = cloud.
`Data Residency`	Edge: process locally, never send PII to cloud. GDPR/HIPAA compliant by design. Cloud: data crosses borders.
`Cost Model`	Edge: per-request pricing (cheap for light compute). Cloud: per-second pricing (cheap for heavy compute). Cost favors running right workloads in right place.
`Offline Capability`	Edge: works without internet. Cloud: internet required. Critical for factories, planes, remote oil rigs, agricultural sensors.
`Deployment Complexity`	Edge: heterogeneous hardware, limited debugging, harder monitoring. Cloud: standardized, mature tooling (CloudWatch, Datadog, etc.).

Pro Tip: Edge computing is about data gravity — move compute to where the data is created or consumed. Not everything belongs in the cloud: if latency must be < 10ms, you're going to the edge.