Mid Sweden University

An increasing actor in the global energy consumption is data centers. It is estimated that by 2025, 20% of the global energy consumption will be caused by data centers and these data centers are continuously running multiple instances of different executables. This study examines the impact of compiler selection on the energy consumption of Java executables, providing insights into how sustainable software development practices can be achieved. The methodology involved compiling five tests from the Computer Language Benchmark Game and one custom test with four different compilers: Oracles Javac, OpenJDKs Javac, Eclipse Compiler for Java, and Janino. The energy consumption of each compiled executable was measured using two different approaches: Running Average Power Limit and an oscilloscope measuring the entire system. Each execution and measuring process was made ten times. Statistical analysis was made using one-way Analysis of Variance to see if there were any significant differences, and confidence intervals were calculated to see if there were any practical differences. The findings show a slight difference in energy consumption between the executables, however, there was no statistical significance or practical difference for most tests. The custom A-poster test displayed statistical significance, but the data had considerable variance detracting from this result's reliability. The findings could suggest that compiler choices have minimal impact on energy consumption, but there could potentially be some savings on a larger scale. However, the results from this study were unreliable, and further research is needed for more reliable data.