Software Energy Measurement at Different Levels of Granularity

Power usage is mainly attributed to hardware. However, hardware resources are controlled by software instructions, which determine how it should behave. This paper presents an overview of the different methods for measuring the power and energy consumption of software programs. We propose a taxonomy in which we classify software measurement methods into different categories from different perspectives. We take into consideration software granularity levels as well as hardware facets. Software granularity concerns the structural facets of software. Hardware granularity concerns the levels of hardware resources. Energy measurements of lower software/hardware levels can be more challenging. We study and evaluate software energy measurement methods for battery-powered devices (e.g., laptops, smartphones, and embedded systems). Our results suggest that some software measurement tools can be capable of generating power readings of lower levels of hardware while some other tools can support estimating power for lower levels of software.