Predição de consumo energético de aplicações OpenMP em máquinas multi-core usando técnicas de regressão de aprendizado de máquina

Abstract

The field of Green Computing research, which aims to make computing more sustainable and environmentally friendly, has been driven by the increasing integration of large-scale data processing and storage technologies. The growing complexity and massive volume of data from various sources has challenged traditional infrastructures, leading to the exploration of multi-core platforms. Despite the significant increase in performance and energy efficiency with the use of multi-core machines, nevertheless, due to the growing demand for consumption, energy expenditure has reached high figures. In this study, we found that polynomial regression models are more effective than linear ones for predicting energy consumption, especially in complex data. In addition, CPU is the largest energy consumer, suggesting the need for optimization or the use of GPUs. We did not find a direct correlation between execution time and energy consumption, suggesting that time-consuming applications may have lower costs due to optimizations. Cluster analysis of the benchmarks indicated similar consumption patterns, useful for future optimizations. Degree 3 polynomial regression was efficient in many cases, but effectiveness varies with the amount of data, and personalized data models proved more efficient than unified approaches.