Skip to main content
eScholarship
Open Access Publications from the University of California

UC San Diego

UC San Diego Electronic Theses and Dissertations bannerUC San Diego

Characterizing and Leveraging Processor Variability in Mobile Devices for Energy Efficiency

Abstract

As semiconductor manufacturers build smaller components, circuits and chips at that scale become less reliable and more expensive to produce, and no longer conform to the rigid hardware specifications usually expected of them. While traditionally, the onus of handling variability has been on the hardware manufacturers, there has been a recent push towards embracing device variability, especially in software, rather than hiding it by increasing guardbands applied to chip designs. Our work takes a higher-level software systems approach to previous studies of embedded sensing systems made in this regard and extends them to mobile devices, mainly smartphones, which are the current generation of general purpose computing devices. We begin by measuring and characterizing processor power variability in mobile devices through fine-grained power measurements on a suitably instrumented platform. We observe variation in processor power consumption ranging from 6% to 15% across smartphones that are manufactured to be identical. This variability, if harnessed properly, could convert into improvements in battery lifetime of 30 to 70 minutes. In this thesis, we also make the case for adaptive software that can leverage information about patterns of variability observed across devices for improved energy efficiency. Using video playback with different tunable parameters as a motivating example, we discuss the trade- off between quality of service and energy usage an the role device variability can play in these trade-offs and find that it is possible for us to proactively choose the encoding and frame resolution parameters to use for video playback, resulting in estimated energy savings of 3-15% which translates to improved battery lifetime of an hour

Main Content
For improved accessibility of PDF content, download the file to your device.
Current View