Adaptive and Flexible Smartphone Power Modeling

Mobile devices have become the main interaction mean between users and the surrounding environment. An indirect measure of this trend is the increasing amount of security threats against mobile devices, which in turn created a demand for protection tools. Protection tools, unfortunately, add an additional burden for the smartphone’s battery power, which is a precious resource. This observation motivates the need for smarter (security) applications, designed and capable of running within adaptive energy goals. Although this problem affects other areas, in the security area this research direction is referred to as “green security”. In general, a fundamental need to the researches toward creating energy-aware applications, consist in having appropriate power models that capture the full dynamic of devices and users. This is not an easy task because of the highly dynamic environment and usage habits. In practice, this goal requires easy mechanisms to measure the power consumption and approaches to create accurate models. The A. A. Nacci · F. Trovò · F. Maggi · M. Ferroni · A. Cazzola · D. Sciuto · M. D. Santambrogio ( ) Politecnico di Milano, Milano, Italy e-mail: [email protected] A. A. Nacci e-mail: [email protected] F. Trovò e-mail: [email protected] F. Maggi e-mail: [email protected] M. Ferroni e-mail: [email protected] A. Cazzola e-mail: [email protected] D. Sciuto e-mail: [email protected] existing approaches that tackle this problem are either not accurate or not applicable in practice due to their limiting requirements. We propose MPower, a power-sensing platform and adaptive power modeling platform for Android mobile devices. The MPower approach creates an adequate and precise knowledge base of the power “behavior” of several different devices and users, which allows us to create better device-centric power models that considers the main hardware components and how they contributed to the overall power consumption. In this paper we consolidate our perspective work on MPower by providing the implementation details and evaluation on 278 users and about 22.5 million power-related data. Also, we explain how MPower is useful in those scenarios where low-power, unobtrusive, accurate power modeling is necessary (e.g., green security applications).