Mobile Performance: Tooling Infrastructure at Facebook
Article Summary
Facebook processes thousands of mobile code changes weekly. How do they catch performance regressions before users feel the pain?
Back in 2015, Facebook engineer Zheng Mi shared how they built CT-Scan, an internal performance monitoring platform that predicts regressions across their mobile apps. The system operates across development, staging, and production to maintain velocity without sacrificing performance.
Key Takeaways
- CT-Scan delivers A/B test feedback in 30-60 minutes during development
- Three-phase approach: development experiments, staging triage, production sampling
- Machine learning predicts issues by analyzing speed, memory, battery, and data usage
- Prevented hundreds of regressions by 2014 using Zookeeper coordination and device labs
- Intelligent testing adjusts frequency based on workload and code impact
Critical Insight
Facebook built a prediction platform that catches mobile performance issues before production by testing across the entire development lifecycle.
