Energy- and Reliability-Aware Task Replication in Safety-Critical Embedded Systems

Abstract

Safety-critical systems should satisfy a required level of reliability. To meet a desired reliability target, task replication can be realized with exploiting multicore platforms. However, inattentive task replication might lead to significant power, energy, and time overhead. In this paper, we demonstrate that when we use task replication technique, the required number of replicas for each task and the energy consumption of the system are significantly dependent on the accuracy of the fault detection. At design time, we propose a method that determines the level of replication along with the voltage and frequency setting for each task to satisfy a desired reliability target such that the energy consumption is minimized. At run time, the proposed method controls cancelling the task replicas in the fault-free scenarios. The proposed method can be applied on both dynamic- and static-priority applications. We evaluated the effectiveness of our method through extensive simulations. The evaluation results show that our proposed method provides up to 43.5% (on average 26.2%) energy saving without reliability degradation.