A model-based energy profiler using online logging for Android applications

This paper describes a model-based energy profiler for Android applications that allows developers to test the energy-efficiency of their applications in real user environments. The profiler offers analysis of the energy consumption using a system-wide power model generated for each type of device. The model, generated by regression analysis, can determine the relationship between power consumption and each behavior of the hardware components. In this paper, we consider two profiler requirements, 1) accurate modeling of up-to-date devices and 2) lightweight online logging on devices for the collection of data. We extend our previous power model to account for the characteristics of multi-core CPUs and 3G/LTE and implement the profiler. Experiments show that it estimates energy consumption with about 10% error in a mixture of applications, while logging incurs a CPU time overhead of only 3.8%, which is superior to other profilers.