The goal of our work was to help developers understand how energy is consumed within their applications and change their applications in ways that will lead to reduced energy consumption. Our work encompassed several related research areas: developing fine-grained source code oriented energy measurement techniques, cataloging developer practices with respect to energy, developing automated optimization techniques, and developing analysis techniques for considering energy as a system quality.

The first area designed methods for source-line level online energy estimation and visualization in ways that allowed us to obtain detailed and fine-grained measurements of how energy is consumed at different levels of implementation and design abstractions.  We developed a technique based on program analysis, power sampling, and statistical methods that enabled energy attribution at a source line level, and also refined an earlier techniques based on execution path based profiling and per-instruction modeling. In evaluation, both techniques demonstrated that they could accurately estimate energy at the source-line level.  The designed algorithms and the evaluation results of both of these efforts resulted in publications that continue to be highly cited at top SE conferences.

The second area used these measurement capabilities to perform systematic empirical studies that correlated implementation structure and energy consumption. This enabled us to understand what affects the energy consumption of software at the different levels of abstraction and led to the identification of a set of best practices for energy-aware software development.  We carried out several large-scale empirical studies into the energy related behavior of mobile apps. These studies led to the identification of several energy-inefficient areas of implementation in app development, mobile ad usage, and database interactions. Results from these studies were published in top-tier software engineering conferences and received national press write ups.

The third area developed automated techniques to help developers make energy-reducing changes to their code.  We developed several static analysis based transformation techniques that could automatically rewrite or restructure apps to reduce their energy consumption.  The techniques that we developed included targeting inefficient designs of graphical user interfaces, inefficiently sequenced network requests, and problematic patterns of writing to a local database.  In evaluation, each of these techniques was able to significantly reduce the energy consumption of the targeted apps. The designed algorithms and the evaluation results of these techniques resulted in publications that continue to be highly cited at top SE conferences.

Finally, in the last area, we considered energy consumption in the broader context of multiple system qualities that must be reasoned about in tandem and that frequently must be traded off with one another. We developed an architecture-based technique for modeling, analyzing, and trading off multiple quality requirements, including energy consumption, and developed a prototype software environment that allows the modeling, analysis, and visualization of key design decisions and their impact on system non-functional properties, including energy consumption.  Finally, we designed an extensible approach and tool for simulation-based analysis of system models that simultaneously considered multiple system qualities. This analysis went beyond state of the art techniques and allowed software engineers to more accurately model app behavior and detect more faults.

Last Modified: 07/26/2018
Modified by: William Halfond