Real-world HTTP benchmarking
summary
Benchmarking web frameworks with TechEmpower, Vert.x developers used perf, async-profiler and JFR with flame graphs and git-watch to locate hot spots. They found heavy CPU time in syscalls from per-request write flushing and improved throughput (≈50–60k → ≈80k req/s) by batching responses; they also implemented a non-blocking PostgreSQL client with HTTP pipelining (level 16) and client-side pipelining to reach top database benchmark ranks.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 8:07Benchmark challenges
- 11:18Network optimizations
- 19:02JIT and inlining
- 27:07Memory allocations
- 29:35Database benchmarks
- 34:32Optimization impact
- 40:13Performance versus ergonomics
- 42:20Multiplexed protocols
- 43:48Conclusion and Q&A
related talks
54mReal-World HTTP Performance Benchmarking, Lessons Learned
33mCrushing Application Latency with Vert.x
38mPractical Performance Analysis
50mReaching 3_000_000 HTTP Requests Per Second
48mFantastic Performance and Where to Find It
01h 03mThe performance loop—A practical guide to profiling and benchmarking
40mPerf Stories
48mBetter performances on the server side with HTTP/2