Energy management of fault-tolerant real-time embedded systems through switching-activity-based techniques

Abstract

Energy management and fault tolerance are essential aspects in designing most real-time embedded systems, and dynamic voltage and frequency scaling (DVFS) is a fundamental approach in energy management. Using DVFS can lead to a considerable decrease in the system's reliability as the rate of transient faults increases due to low voltage and frequency. This study focuses on finding and implementing methods that address the following three issues in hard real-time embedded systems: energy consumption, fault tolerance, and time limitations. Switching activity depends on the nature of tasks. Since the switching activity is task-specific, the energy consumption of a task can be significantly greater than that of other tasks. As far as we know, previous studies on the power management of fault-tolerant real-time systems have not considered switching activity, so the impact of this factor is considered in this study. Multiple heuristic methods are proposed based on the switching activity factor to minimize energy consumption while sustaining the system's original reliability. In the proposed methods, the order of task execution is determined based on task switching activities. The methods are developed for both periodic and frame-based task models. The results show that the proposed methods can decrease the energy consumption by 12.7 %∼35.5 % for the periodic task model and 17.4 %∼45.2 % for the frame-based task model while preserving the system's reliability.