Trash talk
summary
Shows concrete ART changes and measured effects: replacement allocator RosAlloc gives ~4–5× (later ~10–18×) faster raw malloc-style allocation; a large object heap makes 1,000×-size bitmap allocations cost ~0–1 ms instead of 10+ ms; concurrent marking removes one stop-the-world pause and reduces pause time from ~10 ms to ~3 ms; Oreo adds a Concurrent Heap Compaction Collector that runs foreground compaction, cutting system memory by ~30% and turning a 47 s KitKat ray‑trace tile into ~0.1–0.5 s on O.
AI-written summary · about
chapters (10)
- 0:00Introduction to Android memory
- 1:02Dalvik versus ART basics
- 7:35Dalvik garbage collection
- 12:25Introduction to ART features
- 17:06Evolution of ART performance
- 20:10Compaction and bump pointers
- 24:23Modern development advice
- 26:57Allocation demo results
- 30:24Stress testing garbage collection
- 34:42Benchmarks and conclusions
related talks
49mTrash Talk
35mKeynote: Trash Talk: The Evolution of Android Garbage Collection
18mEnd the Senseless Killing: Improving Memory Management for Mobile Operating Systems
29mWhat's new in app performance
38mOptimize your app performance for the next generation of Android
37mUnderstanding Android Runtime (ART) for faster apps
40mAndroid high performance baby
38mLow effort, High impact Android performance wins