Analyzing the Impact of DVFS and System Noise on Merge Sort Benchmarking Through System Quiescing

Abstract

Accurate performance evaluation of computer programs requires reliable execution time measurements. However, modern computer systems contain various hardware and software mechanisms that introduce variability into timing measurements, making the observed performance differ from the algorithm's actual behavior. In this study, the execution time of the Merge Sort algorithm was measured for different input sizes under two experimental conditions: normal system operation and a quiesced system. The results demonstrate that system noise produces noticeable fluctuations in execution time, while reducing these sources of interference significantly improves the stability and repeatability of the measurements. These findings highlight the importance of controlling system noise when conducting performance benchmarking and demonstrate that system quiescing provides a more reliable environment for performance evaluation.