Rethinking age-based garbage collectors

Garbage collection has been around for a long time, and is accepted as a customary element for dynamic memory management in run-time systems of modern programming languages, for reasons of safety and for its clear software engineering benefits. The wide-spread acceptance of object-oriented languages makes it more important than ever to have garbage collection also work as fast as possible. The state of the art in garbage collection on stock hardware are multi-generational copying collectors. We have ourselves developed a flexible (configurable) and efficient language-independent garbage collection toolkit based on this paradigm, and have used it in Smalltalk, Standard ML, and Java. But we recently discovered better ways to collect than generational collection, approaching in some cases optimal performance according to our simulation-based study of programs from object-oriented languages.