Scheduling by Work-Stealing in Hybrid Parallel Architectures
Parallel and Distributed Processing Group (GPPD), Institute of Informatics – Federal University of Rio Grande do Sul (UFRGS), Porto Alegre – RS – Brazil
Federal University of Rio Grande do Sul, 2012
@article{pinto2012scheduling,
title={Scheduling by Work-Stealing in Hybrid Parallel Architectures},
author={Pinto, Vin{i}cius Garcia and Maillard, Nicolas},
year={2012}
}
Nowadays, parallel computing systems have been based on multicore CPUs and specialized coprocessors, such as GPUs, due to the limits achieved by traditional architectures. In order to obtain the expected performance in these systems, the workload must be distributed and redistributed in an efficient way through some technique of scheduling, like work-stealing. This work aims to propose, implement and validate a scheduling approach based on work-stealing in parallel systems with CPUs and GPUs simultaneously. The results show that our approach and Cilk have very close performance. Furthermore, the use of both CPUs and GPUs provides clear improvement in performance due to better utilization of processing resources provided by the workstealing scheduling.
July 24, 2013 by hgpu
