Latency Reduction at the Edge

The Latency Budget Breakdown

 one HTTPS request, no cache hit:

 TCP handshake        1 RTT     ~70ms transatlantic
 TLS handshake       1-2 RTTs   ~140ms (TLS 1.2) 
 request/response     1 RTT     ~70ms
 ─────────────────────────────
 ≈ 280ms BEFORE your server thinks — even with a fast origin!

 edges attack EVERY line of this budget.

Weapon 1: Connection Termination and Reuse

 user ⇄ edge connection established ONCE, reused everywhere:

 user ──(persistent HTTP/2/3, TLS kept warm)──► [edge]
                                                 │ pooled keep-alive
                                                 ▼ connections to origin
 wins:
 - users skip handshake costs after first request
 - edge→origin uses LONG-LIVED pooled connections:
   origin handshake cost paid once per pool slot, amortized
 - mobile clients benefit most (handshakes over radio are brutal)

Weapon 2: Modern Protocols

 HTTP/2 at the edge:
 - multiplexing: many parallel streams on ONE connection
   (no more 6-connection browser limits, no head-of-line queuing
    AT THE APPLICATION LAYER)

 HTTP/3 (QUIC) at the edge:
 - 0-RTT/1-RTT handshakes: resumption in ONE round trip
 - runs on UDP: no TCP slow-start stalls; survives IP changes
   (wifi→cellular handoff without dropping transfers!)
 
 CDNs upgrade protocols to USERS even when origins speak HTTP/1.1 —
 you get modern-client performance for free.

Weapon 3: Route Optimization

 public internet paths are not shortest paths:

 user → origin direct:      congested transit hop, 180ms
 user → edge → optimized CDN backbone → origin: 95ms

 CDNs run private backbones between POPs;
 traffic rides monitored, tuned routes (like anycast +
 steering + peering relationships combined).

 dynamic content benefits too: the MISS path itself gets faster.
 this is why "uncacheable" sites still use CDNs.

Weapon 4: Compression and Shaping at Edge

 - brotli/gzip negotiated per client at edge (CPU cheap there)
 - image format negotiation: serve AVIF/WebP to capable browsers
 - minification/optimization as edge services
 smaller payloads = fewer packets = fewer RTTs on lossy links.

The Composite Effect

 measured typical improvements (far-region users):

 static asset:          280ms → 15ms     (cache + proximity)
 dynamic API call:      300ms → 120ms    (protocol + pooling + routing)
 first-byte TTFB p75:   40-60% reduction typical full-site

 latency compounds through page loads (request chains):
 cutting each hop's RTT multiplies across every dependent fetch.

Interview Framing

“CDN for an API that can’t be cached?” is the trick-probe. Scored answer: yes — termination/pooling kills handshake RTTs, HTTP/2/3 multiplexing helps chatty mobile clients, backbone routing shortens miss paths, compression cuts payload RTTs. Understanding WHY CDNs accelerate dynamic traffic separates candidates who think CDNs are just image caches.