Incremental incrementally compacting garbage collection

Abstract

A mixed-strategy garbage collection algorithm is presented, which combines mark-and-sweep and copy collection. The intent is to benefit from the compacting and linearizing properties of copy collection without losing computational use of half the memory. The stop-and-collect version of the algorithm is a simple and cheap technique to fight memory fragmentation. The collection strategy may be dynamically adapted to minimize the cost of collection, according to the amount of memory actually accessed by the computing process. The parallel version of the algorithm is to our knowledge the only parallel compacting collector for varisized cells, that leaves most (more than half) of the memory available for the computing process.