Leveled Garbage Collection for Automatic Memory Management

Generational garbage collection is one of the most popular memory management techniques. However, poor predictions of object lifetimes can greatly reduce the effectiveness of generational garbage collectors. Some research has addressed this problem, but without much success. This dissertation presents Leveled Garbage Collection, a new algorithm that is not based on object ages. It uses a heap structure and collection scheme similar to those of generational garbage collectors, and has a non-age-based promotion policy that doesn''t promote all of the live objects, but still guarantees ample free space immediately after each garbage collection. Performance evaluation shows that Leveled Garbage Collection not only substantially outperforms generational garbage collectors on a wide range of problems, but also demonstrates superior virtual memory performance.