Generational Garbage Collection Algorithm Based on Lifespan Prediction

Abstract

This paper proposes a new method of generational garbage collection (GC) algorithm for the issue of low efficiency of garbage collection in embedded virtual machine. Through predicting the lifespan of the allocating objects, the objects predicted of long lifespan can be distributed into the old generation directly, and the copy numbers from the young generation to old generation were decreased, thus the executed time of GC is cute down. In the aspect of generational collection, the young generation implements a non-suspended mode, which the objects allocation and GC are done simultaneously, the old generation uses a integrating strategy, combined lazy-buddy algorithm with mark-sweep algorithm, to achieve the quick allocation and recycle. It not only avoids the copy operation on objects, but also controls the amounts of memory fragments. The experimental on v8 JavaScript engine results shows that this algorithm can reduce the GC time by about 23.9%, operation time by around 17.2%, and the overall operation performance of the system is enhanced evidently.