Deploying HTTP/3 at Scale on Mobile: A 17% Startup Latency Gain Despite Implementation Constraints
summary
A randomized controlled rollout across 18 markets shows 51% handshake latency reduction and up to 17% faster p90 home‑screen responses, but some secondary screens regress as much as 14%. Implementation details matter: an OkHttp→Cronet bridge, a Cronet init reduction from ~800ms to ~53ms (~15x), UDP NAT timeouts solved by a keepalive ping, and Cloudflare’s congestion‑control choice (cubic vs BBR) drive surprising throughput and packet‑loss behavior. Follow the measured numbers and vendor settings, not the protocol hype.
AI-written summary · about
chapters (15)
- 0:00HTTP/3 deployment journey
- 1:05Mobile network performance
- 7:04HTTP/2 vs HTTP/3 evolution
- 10:36HTTP/3 performance gains
- 15:34Benchmarking HTTP/3 vs HTTP/2
- 16:37Server-side HTTP/3 steps
- 18:22Mobile HTTP/3 support
- 19:47Cronet with OkHttp integration
- 21:51Optimizing Cronet time
- 23:15QUIC connection timeouts
- 24:18Performance gains impact
- 26:05Regression and congestion control
- 29:19Packet loss impact
- 30:40CDN performance comparison
- 31:25Q&A
related talks
34mSupercharging App Startup with HTTP/3: Lessons from Foodpanda
52mPanel 2: Network
57mHigh Performance Mobile
35mOn QUIC, HTTP/3 and tech stack unification
46mHTTP/3: It's all about the transport!
45mWhy Fast Matters
36mDesigning Apps to Handle Slow Connections
37mAdapting to the Mobile Web F̶u̶t̶u̶r̶e̶ Present