Performance, Computing, and Communications Conference, 2002. 21st IEEE International (2002)
Phoenix, AZ, USA
Apr. 3, 2002 to Apr. 5, 2002
Qian Yang , Dept. of Comput. Sci., Illinois Inst. of Technol., Chicago, IL, USA
The performance issues of garbage collection (GC) have been studied for over three decades. This paper uses a new cycle accurate timing tool to measure GC metrics such as allocation latencies, component elapse time (mark, sweep, and compact) and object life span. The data are then used to derive runtime heap residency and overall GC time. In the past, researchers study object life span through a space based approach, where the amount of allocated memory determines GC invocations. We propose a tune based methodology as a complement. Time plays an important role in server environments, where allocations can come in bursts. The experimental results indicate that a time based approach yields significantly less GC calls, while maintains almost the same heap residency as the space based approach. This translates to a more efficient way to collect garbage.
Q. Yang, W. Srisa-an, J. Chang and T. Skotiniotis, "Java virtual machine timing probes: a study of object life span and garbage collection," Performance, Computing, and Communications Conference, 2002. 21st IEEE International(PCC), Phoenix, AZ, USA, 2002, pp. 73-80.