MathOps or Mathematics in Monitoring
summary
Explains using mathematical and statistical methods in systems monitoring, focusing on metrics like CPU, iowait, load average and service response times. Covers why simple averages mislead for response timings and advocates using medians and high-percentiles (e.g., 99th) while examining what percentiles actually mean and how to apply them in monitoring.
AI-written summary · about
chapters (20)
- 0:00Introduction
- 2:36Queuing systems and metrics
- 6:10Data visualization and charts
- 8:15Normal distribution limits
- 10:01Statistics and percentiles
- 13:32Why mean fails
- 16:44Three sigma and outliers
- 21:17Percentiles in real systems
- 23:26Percentile misuse errors
- 27:20How percentiles are calculated
- 31:35Users vs requests
- 33:03Q&A
- 33:25Threshold selection
- 36:57Choosing metrics correctly
- 38:47Trend visualization and analysis
- 40:36Additional math metrics
- 43:06Product metrics and experiments
- 44:53User feedback
- 46:41Robustness and user metrics
- 48:32Conclusion and answers
related talks
41mMetrics Applied
17mRealtime Monitoring
31mWe Monitor Everything… Except What Matters
36mMetrics - What's your code actually doing?
39mThe Many Ways Your Monitoring Is Lying to You
12mModel Monitoring: Detecting and Analyzing Data Issues
01h 01mFinding Metrics that matter and using them safely
38mKnow your data: The stats behind your alerts