CAREER: Exploiting Instruction-Based and User-Behavior-Based Prediction for Energy and Performance Management in Operating Systems

University Of Arizona, Tucson AZ

Investigators

Abstract

Improving energy efficiency is critical since future users will demand mobile devices with even longer battery life or higher performance stationary systems that are cost effective to operate even when energy costs increase. The challenge that efficient energy management faces is matching the power state of the system resources to the application or user demand. The goal is to reduce energy consumption by keeping the device in the low-power state when not used, but at the same time provide high response when the demand for the device arrives. Current solutions are either not transparent and require programmer involvement, or are less effective in reducing energy consumption and can expose significant delays to the application and the user. The goal of the proposed work is to design a fully autonomous and unified system that will transparently examine and adapt to application demands of both interactive as well as non-interactive applications in the current and future systems. This goal will be achieved through highly detailed user and application behavior monitoring in the operating system for interactive system, and application monitoring for non-interactive server type systems. Capturing and understanding such user and application behavior is critical in meeting user demand for high performance while saving energy. The project will involve students in designing and evaluating energy-efficient mechanisms for various system components and device types through in-class assignments as well as independent studies. The projects and assignments will provide students with better understanding of the challenges that they will face in their future careers.