Power consumption, which has increased dramatically with IC technology advancement and system complexity, has become a critical issue in design of more complicated, versatile, and reliable systems. The International Roadmap for Semiconductors (ITRS) report estimates that a 20-times dynamic power consumption gap and as much as 800-times standby power consumption gap for portable devices need to be closed by the year 2016. Left unchecked, power consumption will curtail the feasibility of future advanced real-time embedded systems. This NSF CAREER research project is leveraging current operating system functionality to support real-time and power-aware computation that can meet the increasingly stringent timing and power/energy constraints for advanced real-time and embedded applications. In particular, this research is developing innovative scheduling techniques and decision functions that can exploit hardware capabilities and make proper tradeoffs. These techniques are based on knowledge of hardware infrastructure features and the intentions of applications, as communicated through the operating system layer. The expected impact of this research lies in its promise to alleviate the inadequacy and inefficiency of power and energy conservation in todays real-time embedded applications. Such applications are of great economic importance. They already prevalent and are becoming increasingly pervasive. Through experimental, application-based research, this project seeks to validate and provide sound studies of its theoretical system-level power reduction techniques under practical scenarios such as automotive electronics. Integration of research and education is carried out under this CAREER project via the course development, graduate/undergraduate student mentoring, out-reach, and dissemination of results. The project offers hands-on learning opportunities for pre-college, undergraduate, and graduate students.
Over the past few decades, we have seen tremendous improvement of real-time embedded systems, thanks largely to the remarkable improvement of semiconductor technology. From cell phones, microwave ovens, automobiles to aircraft control, these systems have the power to profoundly change the way we live our lives. However, as semiconductor technology continues to advance, the rapidly increasing transistor density and the exponentially growing power consumption of IC chips have imposed enormous challenges and threaten to handicap the availability of future generations of computing systems. The goal of this project is therefore to research new resource management methods and techniques that can meet the increasingly stringent power and energy consumption constraints in computing systems. At the same time, the PI also intends to promote the integration of this research into undergraduate and graduate education. The principle of the approach is simple: while many advanced computer systems can deliver high peak performance, practical applications rarely need computer systems to run at their peak performance all the time. The challenge, however, is how to dynamically modulate the computer system performance accordingly without violating the constraints such as timing, power, temperature, etc. Some key results and findings from our research are summarized as follows: While many scheduling techniques focus solely on reducing the processor energy consumption, other components in the systems such as display and storage may also consume extensive power/energy consumption. We developed new scheduling algorithms to reduce the system-wide energy consumption for real-time applications with different QoS guarantee requirements. As the power consumption continues to increase, the leakage/temperature relationship plays an increasingly important role in power/energy aware computing. We studied different models to account for the leakage/temperature dependency and identified an appropriate linear model. By incorporating this linear leakage/temperature model into system level analysis, we developed a number of feasibility checking algorithms, which form the foundation for thermal aware real-time system analysis. Since leakage power depends on temperature and temperature also depends on overall power consumption, the calculation of energy consumption becomes a challenging problem. We developed an analytical method to calculate the energy consumption for a given schedule. This method is both fast and accurate. We also developed a so-called m-oscillating schedule for the thermal sensitive situation. According to this policy, instead of using the high or low speed each for a long interval, we can divide the high speed schedule and low speed schedule into m sections and use high/low speeds alternatively. We have established and formally proved a number of principles and guidelines for thermal aware real-time scheduling. We showed that for a specific interval, a constant speed schedule is NOT necessarily the optimal schedule in reducing the peak temperature. In addition, we showed and formally proved that that for periodic tasks, the peak temperature during the stable status (assuming the initial temperature is the ambient temperature) can be minimized using the constant speed schedule. If such a constant speed is not available, then a two-speed schedule using the two closest neighboring speeds can lead to the lowest peak temperature among the two speed schedules. We developed many power/thermal aware scheduling algorithms to minimize the overall system energy consumption under a peak temperature constraint, to reduce the peak temperature, and to maximize the throughput of a computer system under a peak temperature constraint. We have studied the power/thermal aware scheduling problems based on test beds developed from practical hardware platforms. We have also taken into considerations of the practical limitations in real hardware platforms, incorporated into our theoretical research and developed new scheduling methods. We have also studied a number of scheduling problems including the utility aware scheduling problem, fixed priority semi-partitioning scheduling problem, and non-preemptive EDF scheduling problem. Through the entire course of this project, total 15 graduate students and 17 undergraduate students worked for this project. Three Ph.D students and three M.S students were graduated with degrees. Among the 17 undergraduate students, 13 of them were supported by the REU supplement for the project, and 11 of them are from the underrepresented group (women or minority). Based on research from this project, we have 14 journal papers and 32 peer-reviewed conference papers published/accepted. PI also has developed a new undergraduate level course: "Embedded System Design," and two new graduate level courses: "Real-Time Systems and Applications," and "Power Aware Computing."
