JVM Configuration Parameters Space Exploration for Performance Evaluation of Parallel Applications

Abstract

In the last decade the first parallel processors were launched for desktop computers. Called CMPs (Chip Multiprocessors), they have made parallelism popular, which became the dominant computing paradigm. The high processing capacity and wide range of CMPs models existing has increased the use of virtualization. One of the most used virtual environments is the Java Virtual Machine (JVM). This paper presents an exploration of the JVM configuration parameters space for performance evaluation of parallel applications belonging to the SPECjvm2008 benchmark. The results show that the JVM is unable to implicitly choose their optimal configuration, and it is possible to obtain significant gains through explicit configuration. Experimental results shows that on average, an application can achieve up to 32.9% speedup. The applications were grouped according to the most relevant parameters to its performance, and also for its characterization. It was observed that there is a correlation between the workload characteristics and the relevant parameters to its performance.