I love Big-O but it deceives me
summary
Explains why Big‑O can be misleading in practice because hidden constants, I/O latency and CPU/cache behavior often dominate real-world performance. Demonstrates using pprof and flame graphs for profiling and gives concrete failures: a 10 MB JSON parse that caused an accidental O(n²) slowdown in GTA Online taking minutes (fixed for a $10k bounty), and an internal site suffering 15 s page loads from an N+1 SQL problem that was reduced to ~500 ms by fetching ~5 MB once and doing server-side filtering.
AI-written summary · about
chapters (8)
- 0:00Start
- 8:32The Log
- 11:52Positive Examples
- 14:28Blog
- 16:44Ignore Hidden Coefficients?
- 19:48Performance Tips List
- 20:21Measurements: Profiling
- 37:53Conclusion
related talks
43mI Love Big-O but It Misleads Me
37mCausal's 1000x Spreadsheet; Performance beyond profiling
49mSee the tries for the trees
01h 21mPerformance Really Matters
27mHigh Performance Architecture
47mHigh perf architecture at Stack Overflow
27mLessons Learned From Architecting the Highest-scale Operational Systems in the World
56mWrite Fast Code Like a Native