QUIC, in Theory and Practice
summary
QUIC replaces TCP with UDP while shipping TLS 1.3 and bringing HTTP/2’s multiplexing down a layer to remove TCP head‑of‑line blocking; Google reports QUIC serves ~7% of traffic and shows ~3.6–8% average latency gains and up to 16% at the 99th percentile, YouTube tests show ~20% less rebuffering on lossy links. Implementations live in user space so servers see about 2× CPU cost today and multiple implementations remain immature, so practical wins depend on real‑world deployment and further tuning.
AI-written summary · about
chapters (27)
- 0:00Intro
- 0:57HTTP
- 1:09CP
- 1:54Penpal
- 3:32Packet Loss
- 5:06TCP Headofline Blocking
- 6:12HDP Headofline Blocking
- 6:40Aquaman
- 6:53Parallel Universe
- 7:54Summary
- 10:49HTTP II
- 11:26Index Cards
- 14:52UDP
- 15:26TCP Stack
- 16:07QUIC
- 16:29Why QUIC
- 17:33Security
- 20:25Congestion Control
- 21:05Netinfo API
- 21:56Performance Gain
- 22:27IETF
- 23:31Google Search
- 24:17Mobile
- 26:37User Space
- 27:47Summarize
- 28:26Switching
- 29:32Wonder Woman
related talks
51mDown the line: Evolving HTTP and making things QUIC
19mThe Future With QUIC
40mHTTP/3 - HTTP over QUIC is the next generation
51mHTTP/3 and QUIC: Who, what, where, when and, WHY?
30mThere and back again: reinventing UDP streaming with QUIC
30mHTTP/3 Is Next Generation HTTP. Is It QUIC Enough?
35mOn QUIC, HTTP/3 and tech stack unification
28mHTTP 2.0 and QUIC - protocols of the (near) future