A Fair-Policy Dynamic Scheduling Algorithm for Moldable Gang Tasks on Multicores
Tomoki Shimizu, Hiroki Nishikawa, Xiangbo Kong, Hiroyuki Tomiyama · 2022 11th Mediterranean Conference on Embedded Computing (MECO) · 2022
Parallel computing techniques have attracted attention in multicore platforms for several decades. In order to take advantage of multicore architectures, task scheduling is crucial to choose the most appropriate resources in efficiently performing the tasks. In particular, the recent systems where the functionality becomes complex have been dynamic, and thus scheduling of tasks is one of the challenging issues. In this paper, we develop an algorithm for dynamic scheduling for moldable tasks on multicore embedded systems. We assume that dynamic scheduling considers moldable gang tasks, which can be decomposed into several parallel threads grouped and executed together in a gang manner at the same time as scheduling. In the experimental scenario, our proposed algorithm achieves shorter schedule lengths compared to the state-of-the-art algorithms.