Better performances on the server side with HTTP/2
summary
Shows concrete benchmarks: a frontend server proxies to a backend with 20 ms backend latency and the blocking thread pool server tops out at a 400 requests/sec plateau under HTTP/2 because each proxied request blocks a thread. Switching to HTTP/2 multiplexing raises throughput, and implementing the proxy with Eclipse Vert.x event-loop and multiple reactor threads uses cores efficiently to deliver far higher sustained requests per second than traditional Tomcat/Spring style servers.
AI-written summary · about
chapters (30)
- 0:00Introduction
- 1:06Outline
- 1:38Who am I
- 3:16The bad news
- 3:42How did we improve
- 4:46TCP vs HTTP
- 9:40bottleneck
- 9:57Is it fun
- 12:24HTTP2 is a frame protocol
- 12:54Settings frame
- 13:20HTTP to HTTP
- 15:32Demo
- 18:56Latency
- 19:47Arctic architectures
- 20:49Constraints
- 21:44Use case comparison
- 22:06Benchmark results
- 22:59Limitations
- 23:34HTTP
- 25:29Vertex
- 27:20Polyglot
- 27:40Reactive
- 29:12HTTP2 to
- 30:35Reactor pattern
- 39:53Event loop concurrency
- 40:55HTTP2 event loop
- 41:22HTTP2 event loop comparison
- 42:50Reactive ecosystem
- 43:33Conclusion
- 44:23Questions
