In the literature, numerous works have modeled user activity on smartphones and the effects on battery life. Power-saving modes prolong battery life by saving energy, but application performance is limited as a result. We investigate performance-energy trade-offs of smartphone applications by investigating three strategies: first, we propose an M/M/1 discriminatory processor sharing queue to act as a smartphone server and measure delays of Android applications; secondly, we form a performance-energy trade-off that takes into account cellular radio transfers using an objective cost function incorporating mean delay and power consumption; and thirdly, we build an online HMM to act as a power consumption model that predicts battery life given recent data transfers. For all three strategies, we obtain logged smartphone activity of over 750 users via an open-source smartphone data-collection application. Hence, we obtain three hypotheses from our strategies: first, delay of applications is approximated well using the beta prime distribution; secondly, power consumption increases as mean delay decreases with battery life prolonged if adjustments are made to cellular radio usage; and thirdly, burstiness is captured by HMMs in both data transfers and rates of power consumption.
Information from pubs.doc.ic.ac.uk/tibMswim.