Designing Application Analysis Tools for Cross-Device Energy Consumption Estimation

Designing green and sustainable IoT applications makes energy consumption a key optimization goal of software development. Modern low-energy devices should be driven by energy-aware software. A promising solution to assist developers in this direction is provided by energy estimation tools. In this article, a method of designing flexible energy estimators is proposed. The introduced solution calculates the expected consumption of programs running on different devices and architectures by using synthetic datasets, the popular Valgrind and Pin profiling tools and the well-established Lasso regressor. In contrast to relevant studies, the emphasis is not on the construction of the most accurate tool, but on the characterization of the correlation between the various metrics (features) and energy consumption, on the comparison between predicting methods and on the construction of practical and easy-to-develop tools. The proposed approach is evaluated using the Polybench benchmark suite in widely used ARM-based systems, achieving an R2 score of 0.96, which is comparable to state-of-the-art approaches.