This well-written paper presents a new algorithm for garbage collection of a Java program running across a heterogeneous set of cores (where some cores execute code significantly faster than others).

While it presents a new algorithm, it is written in an accessible manner with several pages of clear introduction building up to the key points. Although I have not been actively involved in garbage collection (GC) development for many years, I found it very easy to follow the text.

The key idea, obvious once stated, is that it is more efficient to run the GC on a fast core when it is at risk of being outpaced by the mutator and on a slow core otherwise. Doing this optimally requires dynamic allocation of the GC thread to the different cores at different times. The authors present a GC algorithm based on these ideas and claim three to 20 percent improvements in time and energy consumption over existing algorithms. Their analysis appears to be sound and detailed. The paper is 24 pages long, including about 40 references mostly dating from 2000 to 2013.