Analyzing power consumption and characterizing user activities on smartwatches: Summary

In this paper, we study real end-user data by releasing a logger application that monitors users while they are using their watches. We collect traces of real smartwatch user activities from 32 users (4 brands, 7 different models) and a total of 70 days of continuous use. We use traces to characterize usage and power consumption while identifying operational bottlenecks. Specifically, we develop a regression-based power model using system-level hardware components and analyze high-level characteristics: general usage behavior, interaction with the battery, total and component-wise power consumption, network activity, and power consumption of frequently run applications. Our models provide insights about the power breakdown among the hardware components. We present 9 major findings and observations including: (1) battery management is important: majority of users charge their devices every 15 hours; (2) screen and CPU consume more than half of the active power: on average, they use 38.5% and 32.6% of the active power, respectively; and (3) application characteristics affect power consumption: downloaded applications (from third party developers) consume up to 4 times more power than builtin applications.