Access Type

Open Access Dissertation

Date of Award

January 2016

Degree Type

Dissertation

Degree Name

Ph.D.

Department

Computer Science

First Advisor

Weisong Shi

Abstract

Mobile devices are developed rapidly and they have been an integrated part of our daily life. With the blooming of Internet of Things, mobile computing will become more and more important. However, the battery drain problem is a critical issue that hurts user experience. High performance devices require more power support, while the battery capacity only increases 5% per year on average. Researchers are working on kinds of energy saving approaches. For examples, hardware components provide different power state to save idle power; operating systems provide power management APIs to better control power dissipation. However, the system energy efficiency is still low that cannot reach users’ expectation.

To improve energy efficiency, we studied how to provide system support for mobile computing in four different aspects. First, we focused on the influence of user behavior on system energy consumption. We monitored and analyzed users’ application usages information. From the results, we built battery prediction model to estimate the battery time based on user behavior and hardware components’ usage. By adjusting user behavior, we can at most double the battery time. To understand why different applications can cause such huge energy difference, we built a power profiler Bugu to figure out where does the power go. Bugu analyzes power and event information for applications, it has high accuracy and low overhead. We analyzed almost 100 mobile applications’ power behavior and several implications are derived to save energy of applications and systems. In addition, to understand the energy behavior of modern hardware architectures, we analyzed the energy consumption and performance of heterogeneous platforms and compared them with homogeneous platforms. The results show that heterogeneous platforms indeed have great potential for energy saving which mostly comes from idle and low workload situations. However, a wrong scheduling decision may cause up to 30% more energy consumption. Scheduling becomes the key point for energy efficient computing. At last, as the increased power density leads to high device temperature, we investigated the thermal management system and developed an ambient temperature aware thermal control policy Falcon. It can save 4.85% total system power and more adaptive in various environments compared with the default approach. Finally, we discussed several potential directions for future research in this field.

Share

COinS