Detecção de Anomalias de Desempenho em Aplicações de Alto Desempenho baseadas em Tarefas em Clusters Híbridos

Abstract : Programming paradigms in High-Performance Computing have been shifting towards task-based models which are capable to more readily adapt to heterogeneous and scalable supercomputers. Detecting performance anomalies in such environments is particularly difficult since it must consider architecture heterogeneity, variability, and the capability to obtain trusted measurements. This work presents a case-study about the detection of anomalies in the execution of the well-known tiled dense Cholesky factorization developed with StarPU. Our experiments have been conducted in a variety of hybrid multi-node platforms to demonstrate how we are capable to detect and highlight performance anomalies.
Liste complète des métadonnées

https://hal.inria.fr/hal-01842038
Contributor : Samuel Thibault <>
Submitted on : Tuesday, July 17, 2018 - 6:33:35 PM
Last modification on : Tuesday, February 12, 2019 - 9:58:47 PM
Document(s) archivé(s) le : Thursday, October 18, 2018 - 4:01:58 PM

File

181587_1.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-01842038, version 1

Citation

Vinicius Pinto, Lucas Mello Schnorr, Arnaud Legrand, Samuel Thibault, Luka Stanisic, et al.. Detecção de Anomalias de Desempenho em Aplicações de Alto Desempenho baseadas em Tarefas em Clusters Híbridos. WPerformance 2018 - 17º Workshop em Desempenho de Sistemas Computacionais e de Comunicação, Jul 2018, Natal, Brazil. pp.1-14, 〈http://natal.uern.br/eventos/csbc2018/?page_id=234〉. 〈hal-01842038〉

Share

Metrics

Record views

293

Files downloads

90