HTTP/3 - HTTP over QUIC is the next generation
summary
Protocol moves HTTP onto QUIC running over UDP and claims faster page loads via independent streams and 0-RTT early data, borrowing TLS 1.3 but requiring new TLS APIs. Practical notes: Google’s experiments inform the design, curl supports draft h3-22, but UDP filtering causes ~3–7% connection failures and server CPU spikes of ~2–3x today. Deployment needs robust fallback logic, better UDP stacks and tooling; benefits are real in lossy networks but broad rollout remains work in progress.
AI-written summary · about
chapters (20)
- 0:00Daniel Stenberg
- 4:30TLS
- 5:11Classic HTTP Network Stack
- 5:31HTTP/1.1
- 11:52Ossification casualties
- 15:09Built on experiences by Google QUIC
- 17:18Improvements
- 19:38Build on top of UDP
- 20:04UDP isn't reliable, QUIC is
- 20:51Streams!
- 21:43Independent streams
- 24:34Stacks: old vs new
- 25:16Feature comparison
- 26:18HTTP/3 is faster
- 28:12This service - over there!
- 29:07Eight HTTP/3 challenges
- 34:38Ship date
- 35:14Implementations
- 39:05Take-aways
- 39:35HTTP/3 Explained
related talks
30mHTTP/3 Is Next Generation HTTP. Is It QUIC Enough?
34mHTTP/3 is coming
37mHTTP/3: why and where are we?
51mHTTP/3 and QUIC: Who, what, where, when and, WHY?
46mHTTP/3: It's all about the transport!
35mOn QUIC, HTTP/3 and tech stack unification
34mSecuring the Next Version of HTTP: How QUIC and HTTP/3 Compare to HTTP/2
30mQUIC, in Theory and Practice