Exploiting Java objects behavior for memory management and optimizations

Abstract

We study the behavior of Java objects in order to explore potential garbage collection optimization opportunities in Java programs. Our study has led to one probable strategy for using heap buffers effectively through obtaining liveness and type information of objects. To allow examination of the Java memory, we implement a heap profiling tool by instrumenting an existing Java virtual machine. The tool records useful object events during program execution, and groups the objects according to their types. We apply the tool to different benchmarks and analyze the results. We also try to characterize the objects' behaviors and suggest that memory occupied by certain objects can be reused by other objects, leading to improved performance of the program. We implement a simple prototype to demonstrate the feasibility of the object reuse approach. © Springer-Verlag 2004.