Two-phase online memory leak detection

Abstract

Memory leaks generated by a running program may exhaust memory resources and degrade system performance. It is hard to reproduce these kinds of leaks, let alone to identify and fix them early in the development phase. Thus online memory leak detection is required. In this paper we propose a two-phase approach to the online memory leak detection problem for Java programs. In phase one (offline analysis), the user executes the investigated program for a while and analyzes its heap with the proposed agent. The agent then summarizes the total occupied bytes of data structures in the heap offline, and the user is then able to select classes whose object instances seem to be potential leaks by examining the total occupied bytes summary. In phase two (online monitoring), to reduce the space and time overhead at runtime, the agent monitors online the objects of selected classes instead of monitoring all objects in the heap. The approach reports the last accessed time and location (including the source file name and the line number) of each leak candidate, with which the user can identify and fix leaks in the program.