Energy minimization for fault tolerant real-time applications on multiprocessor platforms using checkpointing

Abstract

Relentless technology scaling not only dramatically increased the energy consumption of modern processors, it also makes processors less reliable. In this paper, we study the energy minimization problem for real-time applications on multi-processor platforms while tolerating K transient faults using checkpointing. We first introduce an efficient method to determine the checkpointing scheme that minimizes the worst-case response time for a task set that shares the reserved recoveries on a single processor. We then present a fault-tolerant task assignment algorithm to minimize the overall energy. Experimental results show that the proposed algorithm significantly outperforms other related approaches in energy savings.