2018 Embedded Systems Week (ESWeek)

2 / 2018-04-05 22:24:49

Energy-Efficient Temporal Consistency Maintenance on Multicore Cyber-Physical Systems

Real-Time Databases,cyber-physical system,Temporal Consistency,Energy Efficient,Multicore

Scheduling real-time update transactions to maintain data freshness is an important design goal in real-time database research, especially for data-intensive embedded applications. Despite years of active study, related work on RTDBS power management is relatively scarce. How to schedule transactions to maintain data freshness while minimizing the energy consumption is still a blank area of research. In this paper, we propose, to our knowledge, the first solution to the energy-efficient temporal consistency maintenance problem on Dynamic Voltage and Frequency Scaling (DVFS)-capable multicore platforms. We consider the problem of how to minimize the overall total power consumption on multicore, while the temporal consistency of real-time data objects can be maintained. To end this, firstly, we propose an efficient per-CPU DVFS solution to compute a near-optimal constant slowdown speed for the whole update transaction set, under which the transaction set can be scheduled to meet the temporal consistency requirement while resulting in significant energy savings. Next, by adopting the proposed unicore DVFS techniques on each core, we further propose new energy-efficient mapping techniques to explore energy savings for multicores. Finally, extensive simulation experiments are conducted and the results demonstrate the effectiveness of the proposed solutions.