A New Kind of Analytics: Actionable Performance Analysis
summary
Claims a prototype that combines synthetic linear ramps, production monitoring and data instrumentation with unsupervised machine learning to localize bottlenecks. Tests show end-to-end user latency can differ by a factor of seven from server metrics, a linear ramp to 1,000 concurrent users triples response time, the browser component degrades before ~200 users and the app stack at ~300. The system clusters timing variations and surfaces the likely component to fix.
AI-written summary · about
chapters (20)
- 0:00Introduction
- 0:49Performance
- 3:15Data
- 13:28Linear Ramp
- 14:00Response Time
- 14:17Test Results
- 14:56Performance Troubleshooting
- 16:36What we want
- 17:15Recap
- 17:45Leading Indicators
- 18:21Monitoring
- 18:55Prototype
- 19:52How it works
- 21:15Data Analysis
- 21:37Techniques Used
- 22:05Real Data
- 22:18Example
- 22:57Performance Tests
- 28:02Summary
- 28:52Outro
related talks
44mPerformance Analytics Path of Samurai
06mMeasurement is Not Enough: Rise of Performance Analytics
19mOops my app performance sucks... what if observability could be a game changer?
05mSoftware Analytics for Performance Nerds
19mUser-centered observability: load testing, real user monitoring & synthetics
12mPerformance In Context
18mWebsite performance from user perspective
09mPerformance is the first UX challenge to consider