HTTP/3: It's all about the transport!
summary
HTTP/3 keeps HTTP/2 semantics but replaces TCP+TLS with QUIC (a single protocol that folds encryption into transport), cutting the handshake from about eight steps to a three-step QUIC handshake and halving connection setup latency in many mobile scenarios. QUIC eliminates HTTP/2 head‑of‑line blocking and supports connection migration for phones, but brings real challenges: UDP currently costs ~3x kernel CPU vs TCP, middlebox/firewall and reverse‑proxy changes, and partial browser/server support (Chromium, Caddy, Fastly already active).
AI-written summary · about
chapters (27)
- 0:00Introduction
- 1:36HTTP/3 evolution from HTTP/2 over QUIC
- 3:50TCP handshake and latency
- 7:32Latency vs bandwidth examples
- 10:40Latency impact on UX
- 12:24HTTP evolution for latency
- 15:22TLS overhead and QUIC need
- 17:37TCP origins and internet history
- 18:45TCP packet loss handling
- 20:34Why QUIC replaces TCP
- 21:48Network switching with TCP
- 24:04Head-of-line blocking in HTTP/2
- 26:44QUIC advantages over TCP
- 27:08HTTP/3 benefits over QUIC
- 29:04QUIC standardization by Google
- 30:36CDNs adopting QUIC
- 31:48CPU usage: TCP vs UDP
- 32:36Key takeaways and future of HTTP/3
- 34:05Q&A: QUIC without TLS
- 35:24Q&A: QUIC encryption
- 36:08Q&A: QUIC CPU usage
- 37:29Q&A: QUIC packet loss
- 38:38Q&A: QUIC forward error correction
- 39:18Q&A: QUIC packet ordering
- 40:55Q&A: HTTP clients like curl
- 43:05Q&A: Net neutrality with QUIC
- 44:19Q&A: Non-HTTP QUIC apps
related talks
51mHTTP/3 and QUIC: Who, what, where, when and, WHY?
37mHTTP/3: why and where are we?
34mHTTP/3 is coming
30mHTTP/3 Is Next Generation HTTP. Is It QUIC Enough?
35mOn QUIC, HTTP/3 and tech stack unification
40mHTTP/3 - HTTP over QUIC is the next generation
34mSecuring the Next Version of HTTP: How QUIC and HTTP/3 Compare to HTTP/2
34mQUIC and HTTP/3: Too Big to Fail?!