HTTP: the request-response language of the web

TLS sealed the channel. TCP numbered the bytes. Now comes the protocol on top: HTTP — the language browsers and servers speak to request and receive web pages, APIs, images, fonts, and scripts. For thirty years, HTTP/1.1 handled nearly all traffic. Then browsers demanded more.

What HTTP does in one sentence

HTTP (HyperText Transfer Protocol) is the request-response language browsers and servers use to exchange web pages, APIs, images — everything you see in a browser. The client asks, the server answers. That’s it. But how they ask and answer is where the versions diverge.

The restaurant metaphor

Think of HTTP as a waiter taking orders at a restaurant.

HTTP/1.1: One waiter per table, one order at a time per waiter — you wait for each plate before ordering the next. To go faster, the restaurant hires six waiters per customer (six parallel connections). It works, but it’s wasteful.

HTTP/2: One waiter takes ten orders at once and brings dishes back in whatever order the kitchen finishes them. One connection, many requests in flight simultaneously.

HTTP/3: Same as /2, but a delayed dish in the kitchen no longer makes other dishes wait — they come out independently. Built on a different transport (QUIC over UDP) that isolates delays per request.

One scenario end to end

You navigate to a site. The browser does DNS, TCP, TLS, then sends the first HTTP request. The page’s HTML names dozens more files; the browser fires those requests using whatever HTTP version the server agreed to during ALPN (Application Layer Protocol Negotiation — a TLS extension that lets the client and server pick h2, h3, or http/1.1 during the handshake). Modern sites get HTTP/2 or HTTP/3; old origins still see HTTP/1.1.

Bea · Browser the browser to Sven · Origin server the server: “GET /index.html.” Sven replies with the HTML. Bea sees <script src="app.js"> inside and immediately says: “GET /app.js.” In HTTP/1.1, Bea waits for each file to finish on the same connection before asking for the next. In HTTP/2 and /3, Bea fires all the GETs at once and Sven streams them back interleaved.

The core antagonist: head-of-line blocking

Head-of-line (HOL) blocking is the pattern where a slow item at the front of a queue stalls everything behind it. HTTP/1.1 has HOL blocking at the request level — each connection can carry only one in-flight request-response at a time. HTTP/2 fixed HOL blocking at the HTTP layer by multiplexing streams, but TCP underneath still has HOL blocking at the transport layer. HTTP/3 fixed HOL at both layers by using QUIC, which has stream-level loss recovery.