Estimating the energy consumption of a mobile music streaming application using proxy metrics

Abstract: For users of mobile devices an application with high energy consumption may cause the user to remove the application. For a music streaming service, energy consumption is a factor of competition. However, to reduce energy consumption it must be measured. In this study, proxy metrics, for example CPU usage and number of bytes written to memory, are investigated for their suitability as predictors for the energy consumption of a music streaming application on a mobile device. A literature review is conducted to find which metrics have the greatest impact on energy consumption. Further, the literature review is used to identify potential relationships between metrics and energy consumption. A OnePlus 6T using Android 9 as its operating system is rooted and its battery modified to collect data. The data is collected from three test cases for the mobile music streaming application. Memory and network statistics are gathered using the software strace, CPU statistics are collected using data from the proc file system while energy consumption is measured using a power meter. Based on the results from the literature review, linear models are constructed to model the energy consumption. The results show that many of the metrics are highly collinear. From each pair of collinear variables only one variable is kept. The resulting model had a 32.5% better predictive power than an non-optimised model. The best performing model used transmitted network bytes, read and written memory bytes, and user CPU as predictor variables.