On-line reliability-aware dynamic power management for real-time systems

As VLSI technology continues to advance, the aggressive scaling of transistor size not only dramatically increases the power consumption of a processor chip, but also significantly decreases the reliability of the chip. While Dynamic Voltage and Frequency Scaling (DVFS) techniques have been shown very effective in power reduction, they can adversely affect the reliability and stability of computing systems. In this paper, we present an online power management approach to schedule real-time tasks on a single processor platform to minimize the energy consumption without compromising the overall system reliability. The proposed algorithm recycles the reserved redundant resources dynamically and thus can reduce the energy consumption more effectively. As shown in our simulation study, by exploiting the run-time dynamics, we proposed technique can significantly outperform the previous work in energy savings.