Efficient Power Management Schemes for Dual-Processor Fault-Tolerant Systems

To address the increasingly important energy efficiency problem, a Standby-Sparing scheme has been studied previously to reduce the energy consumption in dual-processor fault-tolerant systems. However, the dedicated (spare) processor in the Standby-Sparing scheme needs to execute backup tasks at full speed to preserve system reliability and thus cannot effectively exploit its available slack time. In this paper, based on our prior work of the Preference-Oriented Earliest Deadline (POED) scheduling algorithm, we study an Energy-Efficient FaultTolerant (EEFT) scheme for dual-processor real-time systems, which can effectively exploit the slack time on both processors for better energy savings. Specifically, the primary and backup tasks are first mapped in a mixed manner to both processors. Then, the POED algorithm is used to schedule the mixed tasks on each processor, where primary tasks are executed as soon as possible while backup tasks as late as possible to reduce the overlapped execution of primary and backup copies of the same task and thus reduce energy consumption. The online scheme that considers the additional slack from the cancellation of backup tasks and early completion of primary tasks is further explored. The proposed EEFT schemes are evaluated through extensive simulations. The results show that, compared to the state-of-theart Standby-Sparing scheme, significant (more than 40%) energy savings can be obtained under the POED-based EEFT schemes.