Trustworthy Systems

Understanding and reducing smartphone energy consumption

Authors

Aaron Carroll

CSIRO's Data61, Australia
UNSW, Australia

Abstract

Modern smartphones are increasingly performant and feature-rich, but because they are battery powered, remain highly power-constrained. Energy management is the art and science of maximising battery lifetime, and effectively doing so requires a solid understanding of how a devices uses energy to inform policy and algorithms backed by accurate data. This work addresses each of these issues.

First, we present a detailed power analysis of two smartphones, the Openmoko Freerunner and the Samsung Galaxy S III. We measure power consumption by direct instrumentation at the circuit level by interposing on the power supplies of the individual components, including CPU, RAM, display, GPU, wireless radios, camera, GPS, storage, audio, and environmental sensors. With this instrumentation in place, we produce breakdowns of how energy is distributed under micro-benchmarks and realistic usage scenarios. We also measure two other devices at the whole-system level to validate our earlier results, and to draw conclusions about how smartphone power consumption is changing over time. Additional to the results presented, we also describe a methodology for instrumenting commercial mass-market off-the-shelf devices.

Based on these measurements we observe that peak CPU energy consumption is increasing due to the advent of multi-core processors in the mobile segment. Thus, effective power management of these will be important for battery life on future mobile devices. Such multi-core processors add a new dimension, the number cores active, to the spectrum of available energy management mechanisms.

In the second part of this work we investigate how this mechanism, which we call core offlining, interacts with the well-established technique of dynamic voltage and frequency scaling (DVFS) for minimising power consumption. We find surprising differences in the characteristics of contemporaneous smartphones, specifically in the importance of static power, which we show to be a critical factor in minimising energy consumption. We design and implement medusa, a policy that exploits our findings to integrate core offlining with DVFS in the Linux kernel. We show that despite its simplicity, medusa obtains energy savings that are at least as good as, and often better than, the algorithms that ship on the studied phones, and that it approaches the optimal static algorithm.

BibTeX Entry

  @phdthesis{Carroll:phd,
    address          = {Sydney, Australia},
    author           = {Aaron Carroll},
    keywords         = {Operating systems; Power management; Energy management; Multi-core; DVFS},
    month            = may,
    paperurl         = {https://trustworthy.systems/publications/papers/Carroll%3Aphd.pdf},
    school           = {UNSW},
    title            = {Understanding and Reducing Smartphone Energy Consumption},
    year             = {2017}
  }

Download