Know your data: The stats behind your alerts
summary
Compares arithmetic, geometric and harmonic means with concrete numbers (arithmetic mean 200 rps vs harmonic 150 rps) and shows harmonic mean weights the smallest values, explains median behavior (median stabilizes but can take ~200 points to return after a spike) and gives sampling evidence: head‑based sampling misses 29–40s traces while tail‑based sampling catches long traces. Shows data scale problems (logs 43→92 TB/hr) and Weibull failure math giving ~65% chance of component failure by 300 hours.
AI-written summary · about
chapters (8)
- 0:00Introduction and stats basics
- 3:09Means and central tendency
- 11:36Median and variability
- 17:16Descriptive vs inferential stats
- 21:38Sampling pitfalls and solutions
- 26:01Distributions and failure models
- 31:42Common statistical pitfalls
- 34:28Summary and key takeaways
related talks
41mKnow your data: The stats behind your alerts
32mStatistics for Engineers
18mLook at Your Data
32mThe Subtle Art of Lying with Statistics
43mThe Subtle Science of Misleading with Statistics
32mLearning Statistics Will Save Your Life
52mMonitoring Design Principles
32mSoftware Performance: A Shape Not a Number