Android’s garbage collection (GC) strategy is investigated in this paper in impressive detail. The authors do a good job of explaining how GC works in Dalvik, have ported a significant number of benchmarks, and have collected statistics that appear to be detailed and accurate.
Unfortunately, the paper is rapidly becoming dated. Newer Android devices run Android Runtime (ART) rather than Dalvik. ART’s GC already includes many of the optimizations suggested by this paper (most notably a generational GC). Parts of this paper do remain relevant, such as the authors’ suggestions of system changes to allow for better and more repeatable benchmarking access.
However, most of this paper is specific to Dalvik, and therefore of limited current interest to most readers. The authors are, of course, aware of this problem and, in a final paragraph, propose a follow-up study of the ART GC. I look forward to reading that future paper.