Dynamic scheduling of imprecise-computation tasks on real-time embedded multiprocessors

Adaptability as an emerging design trend for embedded systems has been of increasing interest. As the representative QoS-adaptable application/software model, Imprecise-Computation (IC) gains growing attentions on its ability to provide scalable execution quality by trading-off system resources. Tight coupling between application temporal characteristics and execution quality makes scheduling imperative. While existing QoS-aware scheduling methods are applicable for uniprocessor scenarios, few efforts address the multiprocessor problem. Moreover, significant leakage contribution to energy consumption, due to advanced silicon processing technology, has further complicated the scheduling problem. In this paper, we present a dynamic scheduling algorithm for IC-modeled tasks on real-time embedded multiprocessor systems, aiming at maximizing runtime QoS under timing and energy constraints. Based on the idea of runtime slack reclamation, we firstly analyze the key factors that affect the QoS optimality during slack distribution. The optimality analysis, then, serves as a guideline for a dual-phase slack receiver selection scheme that identifies the best slack receiver candidates. Compared to state-of-the-art dynamic algorithms, our approach can achieve 54.9 more QoS increase with negligible runtime overhead.