Real-World HTTP Performance Benchmarking, Lessons Learned
summary
The talk analyzes real-world HTTP benchmarking using TechEmpower Framework Benchmark results, focusing on Vert.x performance across ~400 frameworks. It details profiling (flame graphs), JVM tuning including JIT inlining and JVM flags, and network fixes such as keep-alive, fewer per-request flushes and HTTP pipelining to improve throughput (up to 6x), and also examines how Postgres query patterns and client connection behavior affect latency and benchmark ranking.
AI-written summary · about
chapters (10)
- 0:00HTTP benchmarking intro
- 3:17Why benchmarks matter
- 8:48Benchmarking vs simulation
- 12:27Benchmark tools and methods
- 19:49Optimizing HTTP requests
- 26:54Analyzing results
- 31:59Concrete use cases
- 36:59Database and protocol benchmarking
- 44:31Lessons learned
- 51:00Conclusion
related talks
46mReal-world HTTP benchmarking
38mPractical Performance Analysis
47mPractical Performance Analysis
48mFantastic Performance and Where to Find It
33mCrushing Application Latency with Vert.x
01h 03mThe performance loop—A practical guide to profiling and benchmarking
50mReaching 3_000_000 HTTP Requests Per Second
38mYour benchmark may not guide a real application performance