PC Based Escape Analysis in the Java Virtual Machine

Abstract—Current computer architectures are multi-threaded and make use of multiple CPU cores. Most garbage collections policies for the Java Virtual Machine include a stop-the-world phase, which means that all threads are suspended. A considerable portion of the execution time of Java programs is spent in these stop-the-world garbage collections. To improve this behavior, a thread-local allocation and garbage collection that only affects single threads, has been proposed. Unfortunately, only objects that are not accessible by other threads ("do not escape") are eligible for this kind of allocation. It is therefore necessary to reliably predict the escaping of objects. The work presented in this paper analyzes the escaping of objects based on the line of code (program counter – PC) the object was allocated at. The results show that on average 60-80% of the objects do not escape and can therefore be locally allocated.

Index Terms—Escape analysis, garbage collection, java virtual machine, memory management.

Manfred Jendrosch is with the Faculty of Computer Science, University of New Brunswick, Fredericton, NB, E3B 5A3, Canada and the Faculty of Computer Science, Bonn-Rhein-Sieg University of Applied Sciences, Sankt Augustin, 53757, Germany (e-mail: manfred.jendrosch@unb.ca).
Gerhard W. Dueck is with the Faculty of Computer Science, University of New Brunswick, Fredericton, NB, E3B 5A3, Canada (e-mail: gdueck@unb.ca).
Charlie Gracie is with IBM Canada Ltd, Kanata, ON, K2V 1C8, Canada (e-mail: charlie_gracie@ca.ibm.com).
André Hinkenjann is with the Faculty of Computer Science, Bonn- Rhein-Sieg University of Applied Sciences, Sankt Augustin, 53757, Germany (e-mail: andre.hinkenjann@h-brs.de).

[PDF]

Cite: Manfred Jendrosch, Gerhard W. Dueck, Charlie Gracie, and André Hinkenjann, "PC Based Escape Analysis in the Java Virtual Machine," Lecture Notes on Software Engineering vol. 2, no. 1, pp. 16-20, 2014.