ERASE: Energy Efficient Task Mapping and Resource Management for Work Stealing Runtimes

Parallel applications often rely on work stealing schedulers in combination with fine-grained tasking to achieve high performance and scalability. However, reducing the total energy consumption context of runtimes is still challenging, particularly when using asymmetric architectures different types CPU cores. A common approach for savings involves dynamic voltage frequency scaling (DVFS) wherein throttling carried out based factors like task parallelism, relations, criticality. This article makes following observations: (i) leveraging DVFS a per-task basis impractical environments cluster/chip-level DVFS; (ii) moldability, single can execute multiple threads/cores via work-sharing, help reduce consumption; (iii) mismatch between tasks assigned resources (i.e., core type number cores) detrimentally impact consumption. In this article, we propose EneRgy Aware SchedulEr (ERASE), an intra-application scheduler top that aims parallel applications. It achieves by guiding scheduling decisions predictions resource configurations. addition, ERASE capable adapting both given static settings externally controlled DVFS. Overall, up 31% improves 44% average, compared state-of-the-art DVFS-based schedulers.