Turning data into leverage
summary
Shows concrete metrics recipes: use median (~400 ms) vs mean (~600 ms) to avoid outlier-driven averages, inspect max (10 s) and 99th percentile (~5 s) to judge tail impact, smooth hundreds of series with a moving average, detect divergence across ~100 servers via median absolute deviation and dynamic thresholds, highlight individual outliers by subtracting per-server median, and compute simple rate of change on ~40 disks to project days remaining and alert a few weeks before full. Sponsored by SignalFx.
AI-written summary · about
chapters (12)
- 0:00Intro
- 0:56Data Source
- 1:27Typical Value
- 1:46Averages
- 2:37Medians
- 3:19Finding culprits
- 4:23Electric Sheep
- 4:41Moving Average
- 6:06Absolute Deviation
- 6:25Mad Deviation
- 7:22Outliers
- 8:27Rate of Change
related talks
05mTurning high-velocity data into leverage for people
41mKnow your data: The stats behind your alerts
37mLies, Damn Lies, and Metrics
18mLook at Your Data
36mMetrics - What's your code actually doing?
52mReal-time-first metrics
44mActionable insight with derived second order monitoring data
06mPerformance is about more than time series charts