Deep dive into the ART runtime
summary
Detailed account of ART internals explains concrete changes: an SSA AOT compiler plus JIT and profile-guided compilation that compiles only ~10–20% of dex code, yielding typical 10x speedups (up to 26x on one Sheets benchmark) and 10–30% faster app startup; a cloud-profile rollout sends optimized profiles at install; GC work moves to concurrent collectors and adds bump-pointer allocation, reducing allocation overhead by ~18x; storage work adds compact DEX and optional uncompressed dex in APKs to save disk and RAM.
AI-written summary · about
chapters (13)
- 0:00Intro
- 2:55Memory or Performance?
- 6:13Major Android Runtime Evolutions
- 8:19Android Distribution
- 9:54ART Optimizations From Dalvik
- 11:24ART Optimizations in Lollipop
- 12:49ART Optimizations in Nougat / Oreo
- 16:55ART Optimizations in Pie
- 17:28Sheets performance
- 18:27Other benchmarks
- 20:13Compiler optimizations
- 25:34Application Lifecycle since Nougat
- 30:35Maps numbers from users
related talks
37mUnderstanding Android Runtime (ART) for faster apps
55mCompilers. Dalvik. ART. And everything in between
34mAndroid or the ART of Optimizing Java Code
39mWhat's new in Android Runtime
16mAndroid ART Optimization in XuanTie C920
29mWhat's new in app performance
35mUnderstanding the Kotlin to Dalvik bytecode translation
19mARTist A Novel Instrumentation Framework for Customizing and Analyzing Apps