Android App Energy Efficiency: The Impact of Language, Runtime, Compiler, and Implementation

With millions of smart phones being purchased every year, the number of mobile applications that have been developed and installed keeps growing substantially. In 2016, Android-based phones dominate the market with 80.7% of market share and there are over 1.6 million apps in Google Play. The battery power limitation requires each app to reduce its impact on the battery life. Existing research and practices focus on how to optimize battery life by reducing network battery drain, judiciously turning off or dimming screens, dynamically adjusting CPU frequency of inactive apps, and changing app behaviors based on different battery states. These studies have greatly helped to improve the energy efficiency of Android apps. However, very few work has been conducted to analyze the impact of programming languages, compilers, runtime, and implementation choices on power consumption of Android apps, which is equally important for Android app developers to write more energy efficient code. This paper addresses the aforementioned problems by making the following contributions: 1) we develop the Android Energy Profiler (AEP) to measure the detailed power consumption of various Android apps, 2) we analyze the impact of different languages (C/C++/Java) and compiler optimization on energy consumption, 3) we study the impact of different Android system runtimes (ART vs. Dalvik) on energy efficiency, 4) we evaluate the impact of implementation choices (recursion vs. iteration and serial vs. parallel) on energy efficiency, and 5) we verify if thin client design can always lead to better energy efficiency.