Experiments for your Android Builds Driven

on YouTube ↗44m

" summary

Uses Gradle Profiler to run reproducible scenarios (example: 2 warmups + 5 measured builds or 4 warmups + 15 measured builds), outputting CSV and HTML reports with means, percentiles and standard deviations. Compares two branches (example JDK8 vs JDK11) using a left-tailed paired t-test (student t) to compute a p value and decide significance (example p ≈ 0 for a clear speedup). Notes automation via CI is possible but slow, and Docker or offline/build-cache flags help isolate environment.

AI-written summary · about

" chapters (9)

1. 0:00Introduction to Gradle Profiler
2. 10:02Measuring Gradle builds with Profiler
3. 14:23Evaluating benchmark results
4. 17:35Statistics for build improvements
5. 21:05Hypothesis testing explained
6. 29:16P-value and statistical significance
7. 31:22Automating experiment analysis
8. 34:18Real-world benchmark examples
9. 37:56Final remarks and Q&A

" related talks

• 43mExperiments for your Android builds driven by Gradle ProfilerUbiratan Soares · Android Summit 2020
• 44mA Healthy Dose of GradlePhilippe Breault · Chicago Roboto 2017
• 37mThe Build ShruggedInaki Villar · Droidcon Vietnam 2019
• 40mImproving Android Build PerformanceVolker Leck · droidCon Berlin 2019
• 37mBecoming and Staying a Productive Developer with Build Scans, Build Validation Scripts and GradleNelson Osacky, Etienne Studer · Droidcon London 2023
• 36mKeeping your team’s Gradle builds in top shapeNelson Osacky, Inaki Villar · droidcon New York 2022
• 37mBuild bigger, better: Gradle for large projectsAurimas Liutikas, Xavier Ducrohet · Google I/O 2019
• 33mIMPROVING ANDROID BUILD PERFORMANCEStefan Oehme · droidcon Berlin 2018