With ever-increasing miniaturization of computers and advances in communication technology, effective energy management in battery-operated embedded systems has gained an unprecedented importance. On the other hand, applications with stringent timing and quality-of-service constraints pose a serious computational burden on these systems. The research community has recently invested significant effort into the investigation of energy management techniques for real-time embedded systems. Nevertheless, most of these studies focus on a single component (e.g. the processor or memory) and a specific power management technique (e.g. reducing the clock frequency or turning off a component). This CAREER research is developing a comprehensive framework for system-wide energy management in real-time embedded systems. The framework includes new formal models, algorithms and tools to reduce overall system energy, across multiple components and through multiple techniques. Another focus of the project is on the lifetime-aware schemes that adjust the performance in terms of the data quality or timeliness, to sustain the system for a given operation (or, mission) time with a fixed energy budget. This research will have a substantial impact on the next-generation of real-time embedded systems by extending their lifetime significantly, and by enabling them to deliver high-quality performance with limited resources.
With the increased pervasiveness and miniaturization of computers, energy management has become a major problem. Improving the system's battery life in the face of increasingly-more-demanding applications (e.g., live audio-video and games) is critical for portable devices. Even for computers that are connected to the power grid, in particular for servers and large Internet data centers, reducing the power consumption and the related heat dissipation have been recognized as prime objectives. Hence designing effective energy management schemes for computer systems has important economic, environmental, and scientific implications. As opposed to the prior studies that focused mostly on a single system component (e.g., the central processor or memory) and a specific energy management technique at the exclusion of others, a system-level energy management approach has been adopted in this project. In other words, the project considered the goal of reducing the energy consumption of the overall computer system. Specifically, for real-time embedded computer systems that have to perform also in a timely manner during execution, the project has had the following outcomes related to the intellectual merit. The interplay of existing power management techniques for real-time computer systems has been investigated and quantified. This paved the way for developing and evaluating novel energy management schemes that explicitly model and target the aggregate energy impact on the entire computer system. The newly developed technique is experimentally shown to outperform other techniques by significant margins, while still meeting operational constraints. Moreover, these results have been extended to emerging computer systems that consist of multiple processors on the same chip (known as multicore systems). In addition, energy management techniques for computer systems that are able to harvest energy from the environment (e.g., networks of wireless sensors), have been devised. As for broader impacts, the techniques developed during the project have the potential of reducing the energy consumption of computer systems, translating to immediate financial and environmental benefits. In addition, the project provided a training platform for a number of graduate students on cutting-edge research problems, preparing them for future advanced research and development positions, and contributing to the goal of enhancing the US scientific and technological workforce.
