Approximation-aware scheduling on heterogeneous multi-core architectures

The high performance demand of embedded systems along with restrictive thermal design power (TDP) constraint have lead to the emergence of the heterogenous multi-core architectures, where cores with the same instruction-set architecture but different power-performance characteristics provide new opportunities for energy-efficient computing. Heterogeneity introduces challenges in scheduling the tasks to the appropriate cores and selecting the frequency assignment of each core. In this paper, we introduce an approximation-aware scheduling framework for soft real-time tasks on the heterogeneous multi-core architectures. We consider multiple versions of a task obtained by introducing approximation in the computation to provide different levels of quality of service (QoS) versus performance tradeoffs. The additional choice of approximation allows us more flexibility in meeting the performance and TDP constraints while maximizing QoS per unit of energy.