This project targets at solving a longstanding, hard problem in operating system research: how to determine the contribution by a process to the system energy consumption when multiple processes are active. Energy is an important system resource for mobile and embedded systems that are battery-powered and thermally constrained. Per-process energy accounting is important for software evaluation and optimization as well as system energy management and security.
This project tackles this hard problem with two novel methodologies. First, it leverages the smart battery interface available in modern mobile systems and statistical learning to estimate system energy consumption for very short time intervals. Second, the project exploits and extends technologies from multi-player game theory, in particular, Shapley value, to distribute the system energy consumption to concurrently running processes. The project targets three novel contributions: (1) system and algorithmic solutions to achieve high accuracy in system energy estimation for very short time intervals without any external assistance; (2) extended multi-player game theory based on Shapley value to distribute system energy consumption to running processes with incomplete measurement; and (3) utility-based system energy management based on per-process energy accounting. Results from this project apply to not only mobile systems but also servers and data centers. By applying statistical learning and game theory to systems research, the project produces novel, interdisciplinary curriculum modules for teaching both these theories and operating systems. The project also generates open-source tools and data sets for energy modeling and optimization in mobile systems.
