Virtual Machine-Assisted Collaborative Junk Object Detection
Memory leak is unrecoverable software bug that causes performance degradation and re- liability issues to software applications. Although memory management systems exist in modern Object Oriented Language to reclaim unused memory store, memory leak can still happen, and continue to exhaust memory resource. In Java, where there is garbage col- lection for releasing unused objects, memory leaks manifest itself in the form of unused object retention. Since Java Programming language allocates objects on heap, the lifetime of an object deviates from the stack discipline, which can be a challenge in detecting Java memory leak. In this thesis, we propose a collaborative approach in detecting Java memory leaks through verifying Object Lifetime Specification at runtime. We designed a runtime verifier that leverages Java Virtual Machine technologies to monitor and extract annotated infor- mation from the user application, and use that information to verify against Java Virtual Machine events to detect unintentional object retention in the Java application under test. We implemented our runtime verifier with Maxine Virtual Machine, an open source, meta-circular virtual machine developed by Oracle Lab, and conducted experiments and DaCapo benchmark to evaluate its accuracy and performance efficiency. The results show that the runtime verification tool successfully identifies junk objects for different semantic cases proposed in this thesis with certain runtime overhead. Through the research and experimental results, we further make implications on how to improve the performance overhead associated with current design and implementation methods in detecting unused object retention, which in the long term constitute memory leak and performance bug.
Cite this version of the work
Ming Ma (2014). Virtual Machine-Assisted Collaborative Junk Object Detection. UWSpace. http://hdl.handle.net/10012/8646