Edge Functions: Running Code at the CDN Layer

Vercel's 1ms Response Times

Vercel serves Next.js applications from edge functions running in 30+ regions. A user in Mumbai gets their personalized page from a function running in Mumbai - not from a US-East server. The response time is often under 10ms, which feels instant. The function runs AT the CDN node, eliminating the round trip to a central origin entirely.

What Edge Functions Can Do

Edge Function Use Cases:

Use Case              Without Edge         With Edge
--------------------  ------------------   ------------------
A/B testing           Origin decides       Edge rewrites URL
                      (200ms+ round trip)  (5ms, no origin)

Auth verification     Origin validates     Edge verifies JWT
                      JWT (200ms+)         (5ms, blocks bad)

Geo-routing           Client-side detect   Edge reads IP geo
                      (slow, inaccurate)   (instant, accurate)

Image resize          Pre-generate all     Resize on-the-fly
                      sizes (storage cost) (compute at edge)

Bot detection         Origin processes     Edge challenges
                      bot traffic (waste)  bots before origin

Localization          Origin renders       Edge substitutes
                      per-locale page      content by header

V8 Isolates vs Containers

Cold Start Comparison:

Approach        Cold Start   Memory    Isolation
--------------  ----------   --------  ----------------
Full container  100-500ms    128MB+    Process-level
(Lambda@Edge)                          (strong)

V8 Isolate      < 1ms        ~2MB      V8 sandbox
(CF Workers)                           (Chrome tabs)

Why V8 Isolates are faster:
  Container: boot OS -> load runtime -> load code -> run
  V8 Isolate: load V8 -> run (same engine as Chrome)

  At 100,000 req/s, cold starts matter enormously.
  1ms cold start: imperceptible
  500ms cold start: visible to users on first request

  Trade-off: V8 Isolates have
  - No file system access
  - Limited CPU time (10-50ms)
  - No native code/C extensions
  - JavaScript/TypeScript/WASM only

Edge Platforms Compared

Platform           PoPs    Runtime       Cold Start  Limits
-----------------  ------  -----------   ----------  -----------
Cloudflare Workers 310+    V8 isolate    < 1ms       10ms CPU
Vercel Edge Func   30+     V8 isolate    < 5ms       30s timeout
AWS Lambda@Edge    200+    Container     ~100ms      30s/5MB
Deno Deploy        35+     V8 isolate    < 1ms       50ms CPU
Fastly Compute     80+     WASM          < 1ms       Custom

Choose Cloudflare: most PoPs, cheapest, fastest cold start
Choose Lambda@Edge: already on AWS, need full Lambda features
Choose Vercel: Next.js app, integrated deployment

Interview Tip

When discussing latency optimization, say: 'For logic that does not need database access - like authentication, A/B testing, geo-routing, or bot detection - I would move it to edge functions running at CDN PoPs worldwide. This eliminates the origin round trip, giving sub-10ms response times globally. I would use Cloudflare Workers for their V8 isolate architecture which has sub-millisecond cold starts, compared to Lambda@Edge containers which can add 100ms+ on cold starts.'

<

/blockquote>

Key Takeaway

Edge functions execute code at CDN nodes globally, enabling sub-10ms response times for logic that previously required an origin round-trip. Use them for A/B testing, auth, routing, and personalization. Cloudflare Workers (V8 isolates) offer better cold start performance than Lambda@Edge.