Run-time Energy Adaptation of Smartphone Software

Smartphones are a new computing platform that requires new software solutions. Limited battery energy requires energy efficient software to maintain acceptable battery life. Unparalleled heterogeneity is present on the platform due to multiple manufacturers, hardware configurations, and operating system versions currently in use. Usage patterns are diverse and complicate energy efficiency by causing wide variations in the energy availability for apps. New software techniques are needed to overcome these challenges to make smartphone software energy efficient.;In this dissertation we present a system framework for online adaptation of smartphone software. Our solution allows software tasks to dynamically select configurations automatically at run-time to satisfy energy constraints. This approach enables smartphone software to maintain energy efficiency despite platform heterogeneity and usage pattern variations. First, we describe how accurate run-time smartphone power measurements can be implemented using built-in battery monitor units. Second, we present a Markov Decision Process (MDP) for adapting delay tolerant tasks. Third, we propose a systematic statistically rigorous Design of Experiments (DoE) based technique for surveying an unknown configuration space. Lastly, we extend our DoE survey with supplemental heuristic searching in a methodology called MELOADES to satisfy additional energy constraints.;Each of these framework components is demonstrated on software tasks that required run-time energy adaptation. MDP is evaluated by optimizing the synchronization rate of Twitter and Email tasks while preserving energy for stochastic phone call arrivals. A camera trap task requiring image capture and wireless transmission is demonstrated using DoE for configuration selection. Lastly, MELOADES further addresses a more general image capture and transmission task to satisfy varying energy constraints.;We demonstrate the ability of run-time adaptation techniques to overcome heterogeneity on smartphones and enable the same software to remain energy efficient across multiple devices.