HTTP/1.1, HTTP/2 and HTTP/3

Comparison

Why multiplexing matters

HTTP/1.1 sends one request at a time per connection, so browsers open about six connections per host and queue the rest. HTTP/2 sends many requests in parallel as interleaved streams over one connection.

But HTTP/2 still runs on TCP: one lost packet stalls every stream until it's retransmitted (TCP head-of-line blocking). HTTP/3 uses QUIC, where each stream is delivered independently, and connection setup and TLS happen together in one round trip. QUIC connections also survive a change of network (Wi-Fi to mobile).

import math

rtt_ms, requests = 100, 30

http1 = math.ceil(requests / 6) * rtt_ms    # 6 parallel connections, one request each at a time
http2 = rtt_ms                              # all 30 multiplexed on one connection
print(f"HTTP/1.1 with 6 connections: {http1} ms of request round trips")
print(f"HTTP/2 multiplexed:          {http2} ms")
print("(plus connection setup: TCP + TLS 1.3 = 2 RTT for HTTP/1.1 and HTTP/2, 1 RTT for HTTP/3)")

HTTP/1.1 with 6 connections: 500 ms of request round trips
HTTP/2 multiplexed:          100 ms
(plus connection setup: TCP + TLS 1.3 = 2 RTT for HTTP/1.1 and HTTP/2, 1 RTT for HTTP/3)

What it means for you

• Put HTTP/2 or HTTP/3 on your load balancer or CDN; backends often still speak HTTP/1.1 behind it.
• Old HTTP/1.1 tricks such as domain sharding and sprite sheets are unnecessary with HTTP/2.
• gRPC requires HTTP/2 end to end, so your load balancer must support it.