Dynamic voltage and frequency scaling (DVFS)technique is emerging in various battery-operated embedded systems to reduce the energy consumption and prolong the working life of the system. However, DVFS technology has been proved to have some direct and negative effects on the reliability of the system. Most existing schedulers of real-time tasks based on DVFS only focus on minimizing energy consumption without taking the fault-tolerant into account. To solve this problem, in this study, we developed a novel Energy-Aware Fault-Tolerant (EAFT) technique that was tailored for the real-time periodic tasks. The heuristic EAFT balances the allocation of slacks used for reducing energy consumption and used for reexecuting the failed tasks. The simulation results showed that the proposed reliability-aware schemes could guarantee the system reliability and significantly save energy comparing to the existing allocating schemes.
BackgroundWith the advance of pervasive computing and ongoing miniaturization of computers, battery-operated embedded systems have been widely applied into many real-time systems. In such systems, energy has been promoted to be the first-class system resource and the energy-aware system design has recently become an important research area. The DVFS technique has been frequently used to exploit unused CPU time by scaling down processor frequency and supply voltage simultaneously in an effort to save energy.Numerous studies over the past decades have been conducted to reduce the overall energy consumption for real-time tasks scheduling in embedded systems. The work by Weiser et al.[1] was among the first to propose scheduler for non real-time tasks based on DVFS. Yao et al.[2] provided a static, optimal, and polynomial-time scheduling algorithm for real-time aperiodic tasks with worst-case execution times. Most of these studies focused solely on reducing the energy consumption. However, reduced voltage easily leads to occurring of transient faults [3,4]. On the other hand, it is well known that real-time task always has strong time constraint and fault-tolerant requirements. Although fault-tolerance has been studied extensively too, fault-tolerance and DVFS researches have remained somewhat isolated to each other. So far, only a few studies investigate scheduling to meet both faulttolerance and efficient energy requirements. Dakai et al proposed and SUEF-based[6] scheduling algorithms to implement the balance faulttolerance and energy-saving.The aforementioned existing scheduling algorithms mainly assumed that operating frequency could be scaled continuously arbitrarily. Nevertheless, practical commercial processors only support several discrete levels of supply voltages and frequencies. To fill this gap in real-time technology, in this study, several discrete supply voltages and operating frequency levels and a set of real-time tasks, with different periods, are taken into consideration. Compensating for the negative effect on reliability, we adopt the primary/backup fault-tolerant mode. Basing on ...