| Nowadays, exponentially increasing demands for portable devices and more sophisticated and intelligent functionalities in embedded applications make processors much power-hungrier. Due to a limited energy budget, low-power computing at a relatively low level of power consumption or power-aware computing with the knowledge of power source plays important role in extending systems' lifetime. In the dissertation, we study how to reduce power consumption using a low power voltage-clock scaling scheme in real-time systems that require strict time and energy constraints. The property of power-delay tradeoff has significant impact on the schedulability as well as on the performance of the systems.; First, we focus on dynamic reclaiming of early released resources in earliest deadline first (EDF) scheduling using voltage-clock scaling. In addition to a static voltage assignment, we propose a new dynamic-mode assignment, which has a flexible voltage mode setting at run-time, enabling much larger energy savings. Using simulation results and exploiting the interplay among power supply voltage, frequency, and circuit delay in CMOS technology, we find the optimal two-level voltage settings that minimize energy consumption. Next, we study battery-driven real-time systems that jointly schedule hard periodic tasks and aperiodic (sporadic) tasks, whose power sources are bounded in a feasible range decided by a set of tasks. Therefore, reducing power consumption is not the only objective of task scheduling in this case. To make the most of the available energy budget, an effective energy-sharing scheme is proposed using two-mode voltage-clock scaling-EDF scheduling.; Based on the proposed energy allocation model for energy-constrained real-time systems, a dual-policy dynamic scheduling method is proposed not only for faster responsiveness but also for reducing power consumption. The feature of the approach is an intermixing of two schedules; the one consumes minimum energy and the other the worst-case energy, respectively, leading to much longer lifetime under a bounded energy budget. Lastly, we build a scaling mechanism that optimizes energy saving and responsiveness to the aim of the system's performance. Fully utilizing the property in sharing constrained-energy and processor's capacity in total bandwidth server, we adjust total power consumption and average response time of aperiodic tasks by simply controlling the utilization of periodic tasks. |
