A Performance Evaluation Platform for Cyclic Reference Counting

• Main Authors: Jia-Han Yang, 楊佳翰
• Other Authors: Ting-Wei Hou
• Format: Others
• Language: zh-TW
• Published: 2006
• Online Access: http://ndltd.ncl.edu.tw/handle/87799748388001694101

碩士 === 國立成功大學 === 工程科學系碩博士班 === 94 === With the increasing incorporation of garbage collection runtimes into high-level programming languages, such as Java and C#, many researchers have been reconsidering garbage collection (GC) algorithms. However, the evaluation of GC algorithms often involves implementing the algorithm on a specific environment, for example, a virtual machine. This is especially difficult for researchers who are not familiar with the environment because it would take a considerable time to learn. A performance evaluation platform is designed and implemented for GC algorithms. The main idea is to offer an evaluation environment, which is system-independent, modular and easy to use, so that the burden of GC researchers can be relieved significantly. This platform is preliminarily designed only for Cyclic Reference Counting (CRC), a well-known issue for Reference Counting. Key to this design is that it takes a “real” computation graph for CRC algorithms. All the input (graphs) are obtained by taking snapshots of memory when running a known CRC algorithm on the Jikes Research Virtual Machine (RVM), in which the SPECjvm98 benchmarks are used. The evaluation is thus close to the real situation. By this platform, one can save much time when developing or evaluating a CRC algorithm. In addition, it also helps developers to determine whether to implement the algorithm in a real system or not, such as a virtual machine.