Edge Computing / Edge Functions

edge computing · edge functions · edge compute · cloudflare workers · vercel edge · lambda at edge · edge runtime

Definition

Edge computing is the execution of code on CDN nodes near the visitor; in e-commerce it is used for personalization, A/B testing and content control with minimal delay and no call to a central server.

What edge computing is

In the classic architecture the browser requests a page, the request travels to a data centre that may be thousands of kilometres away, the server builds a response and the response comes back. Even with a fast server, physical distance costs time.

Edge computing breaks that pattern. The code runs on the CDN node nearest the visitor. A request from Berlin is handled at the Berlin point of presence instead of flying to Amsterdam or Singapore.

Classic: Browser → CDN (cache) → Server → Response   ~100–300 ms
Edge:    Browser → Edge function (CDN) → Response    ~1–10 ms

Platforms and runtimes

Platform	Runtime	Notes
Cloudflare Workers	V8 isolates	The lowest latency, 200+ points of presence
Vercel Edge Functions	V8 / Next.js	Integrated with Next.js middleware
AWS Lambda@Edge	Node.js / Python	Flexible, but higher latency
Fastly Compute	WebAssembly	Any language, through WASM

Edge computing and personalization

For e-commerce, edge functions earn their place in three scenarios:

A/B testing with no content flicker. The split between variants happens at the edge before the HTML is served. There is no JavaScript swap after load, so there is no flash of the wrong content.

Geotargeting and redirects. Working out the visitor’s country and sending them to the right language domain takes 1–2 ms at the edge, against 50–100 ms with a trip to the origin.

Personalized titles and meta tags. Changing the title, description or og:image by segment to improve click-through from social platforms.

Important: heavy ML work — recommendations, ranking — does not run at the edge. The edge function only collects an already computed result from the CDN cache or a key-value store and drops it into the response. The model itself runs asynchronously in a cloud ML service.

Limitations

A restricted runtime: no full Node.js API, no direct database access
Cold starts: the first request to a new isolate costs another 1–5 ms
Debugging is harder — local emulation of an edge runtime is not always identical to production
CPU and memory caps: bulk computation does not belong at the edge

Frequently asked questions

How do edge functions differ from ordinary server functions?

An ordinary server function runs in one data centre. An edge function is replicated to hundreds of CDN nodes worldwide and executes on the one nearest the visitor. Latency drops from 50–300 ms to 1–10 ms. The trade-off is that an edge runtime offers fewer APIs — no file system, a reduced Node.js surface — and strict execution time limits.

When is edge computing worth using for personalization?

When personalization is performance-critical: A/B testing without flicker, redirects by geography, substituting geotargeted content. It is not the place for heavy ML work — that needs a full API server, and only the precomputed result is passed to the edge.

Does edge computing work alongside a server-side platform integration?

Yes. The typical architecture has the edge function fetching precomputed recommendations or segment parameters from an external API backed by an ultra-low-latency CDN cache, inserting them into the response and returning it. The heavy ML inference itself runs asynchronously in the cloud backend.

Are there data size limits in edge functions?

Yes. Cloudflare Workers allow 128 MB of memory and a maximum of 50 ms of CPU. Vercel Edge allows 128 MB with no hard timeout. For personalization that is enough: a full ML index cannot live at the edge, so the data is normally read from a key-value store such as Cloudflare KV or Vercel Edge Config.