Memory Management on Mobile Devices

on YouTube ↗20m

" summary

Evaluates Android Runtime garbage collection with a reproducible framework, curated 10 real Android apps benchmarked via UI Automator, and compares DaCapo plus a microbenchmark; implements a heap-size hijack via zygote specialization so individual apps run with controlled heaps without rebooting; measures empirical lower bounds on GC overheads and reports they range from 2% to 51% for modest heaps; finds semispace vs concurrent copying tradeoffs and shows some apps (Twitter) give chaotic results due to dynamic content.

AI-written summary · about

" chapters (12)

1. 0:00Introduction
2. 1:04Motivation for mobile memory
3. 3:49Android runtime challenges
4. 5:14Measuring GC cost with ALOW
5. 7:00Overcoming heap limits in Android
6. 8:25Heap introspection issues
7. 9:49Benchmark suite for Android apps
8. 13:22Experimental setup
9. 14:02Vanilla Java results
10. 14:53Android app results incl. Twitter
11. 16:30Conclusion
12. 17:12Q&A