Load Balancing for Iterative Applications in Heterogeneous Architectures

• Guilherme Galante UNIOESTE
• Luis Trivelatto UNIOESTE
• Edmar Bellorini UNIOESTE

Resumo

The objective of this paper is to present a multilevel load balancing library for iterative applications, considering its execution in clusters of heterogeneous nodes, composed of multiple cores and accelerators. The main contribution of this research is the possibility of performing the load balancing between the heterogeneous nodes of the cluster and also in a second level, redistributing the load between the processor cores and accelerators. The library proved to be effective, allowing a higher architecture utilization rate and reducing the execution time of the applications.

Palavras-chave: Graphics processing units, Computational modeling, Random access memory, Load management, Computer architecture, Jacobian matrices, Libraries, Load balancing, iterative applications, parallel applications, heterogeneous architectures