About the Event
Smartphones with cellular data access have become increasingly popular with the wide
variety of mobile applications. However, the performance and power footprint of these
mobile applications are not well-understood, and due to the unawareness of the cellular
specific characteristics, many of these applications are causing inefficient radio resource
and device energy usage. In this dissertation, we aim at providing a suite of systematic
methodology and tools to better understand the performance and power characteristics of
cellular networks (3G and the new LTE 4G networks) and the mobile applications relying
upon, and to optimize the mobile application design based on this understanding.
We have built the MobiPerf tool to understand the characteristics of cellular networks.
With this knowledge, we also make microscopic analysis on smartphone application performance
via controlled experiments and macroscopic analysis via the close examination
of a large-scale data set from one major U.S. cellular carrier. To understand the power footprint
of mobile applications, we have derived comprehensive power models for different
network types and characterize radio energy usage of various smartphone applications via
both controlled experiments and 7-month-long traces collected from 20 real users. Specifically,
we characterize the radio and energy impact of the network traffic generated when
the phone screen is off and propose the screen-aware traffic optimization. In additions to
shedding light to the mobile application design throughout our characterization analysis,
we further design and implement a real optimization system RadioProphet, which uses
historical traffic features to make predictions and intelligently deallocate radio resource for
improved radio and energy efficiency.