Tracking Regressions and Improving UI Performance using Benchmarking Approach
summary
Benchmarks measure actual UI costs (create/bind/layout or Litho lifecycle: create/layout/mount) and catch regressions that tests miss: a newsfeed header moves from 10→14 ms (40% regression) and bisecting isolates the culprit commit (bisect to commit 41). Tests run on fleets of real rooted devices, report hourly to a dashboard, and use controls like CPU frequency lock, RAM-mounted disks and network proxies to reduce noise; regressions auto-create tasks and assign blame for fix or revert.
AI-written summary · about
chapters (14)
- 0:00Introduction and speaker
- 1:14Why benchmarking matters
- 5:30UI performance impact
- 8:42Debugging UI issues
- 12:39Need for automated benchmarks
- 13:23Benchmarking example
- 17:40Writing benchmark tests
- 24:45Benchmarking declarative UI
- 27:15Facebook’s benchmarking setup
- 29:02Mobile app benchmark flow
- 30:48Dashboard and regression tracking
- 31:28Benchmarking challenges
- 34:20Gotchas and limitations
- 36:51Q&A
related talks
47mUsing Benchmarks for Tracking Regressions and Improving UI performance
29mHow to Performance
44mAndroid Benchmarking and other stories
48mRevisiting Performance Budgets
38mYour benchmark may not guide a real application performance
17mWhat I Learned Benchmarking React
39mMicro Benchmarks
17mBeyond the Basics: Performance Monitoring and User Experience for Mobile App Growth