Workload scheduling analysis in geophysical numerical methods
The tasks organization among computational nodes affects the performance of the program. In computationally expensive applications such as geophysical problems, the impact is more significant. The imbalance caused by inefficient task scheduling can generate an application non-scalable. Therefore thi...
Na minha lista:
Autor principal: | |
---|---|
Outros Autores: | |
Formato: | Dissertação |
Idioma: | pt_BR |
Publicado em: |
Universidade Federal do Rio Grande do Norte
|
Assuntos: | |
Endereço do item: | https://repositorio.ufrn.br/handle/123456789/30253 |
Tags: |
Adicionar Tag
Sem tags, seja o primeiro a adicionar uma tag!
|
id |
ri-123456789-30253 |
---|---|
record_format |
dspace |
institution |
Repositório Institucional |
collection |
RI - UFRN |
language |
pt_BR |
topic |
Escalonamento de tarefas FWI LSM Work-stealing Centralizado dinâmico Descentralizado estático Descentralizado dinâmico Computação de alto desempenho MPI |
spellingShingle |
Escalonamento de tarefas FWI LSM Work-stealing Centralizado dinâmico Descentralizado estático Descentralizado dinâmico Computação de alto desempenho MPI Santana, Carla dos Santos Workload scheduling analysis in geophysical numerical methods |
description |
The tasks organization among computational nodes affects the performance of the program. In computationally expensive applications such as geophysical problems, the impact is more significant. The imbalance caused by inefficient task scheduling can generate an application non-scalable. Therefore this work investigates the workload scheduling in geophysical methods. Three types of workload scheduling will be compared: centralized dynamic (CD),
decentralized static (DS), and decentralized dynamic (DD). The CD was implemented
with the technique master-slave, where the master node is responsible for the distribution
of the tasks to other nodes called slaves. The DS was implemented with an equal division
of tasks before the execution of them. The DD implements the work-stealing method
proposed by Assis et al. (2019), where an idle node can steal the tasks of an overloaded
node. The principal geophysical method used was 2D Full waveform inversion (FWI) with
the acoustic wave. To analyze the performance of workload scheduling methods, we
employ a synthetic velocity model and present speedup, efficiency, and load distribution
plots generated with different model sizes and different quantity of nodes. The FWI code
and the workload scheduling methods were implemented in C with distributed memory
parallelization and using the message passing interface (MPI) library. With the results of the workload scheduling methods in FWI, it was applied the workstealing (because this technique presented the more effective performance) in another
geophysical problem: Least-squares migration (LSM). We used the LSM with DS implemented by Chauris and Cocher (2017) to compare with the LSM with work-stealing.
The LSM code and DS used in this problem were implemented in Fortran and the workstealing in C. The communication between the nodes was implemented using MPI. To
analyze the performance of workload scheduling methods in LSM, we used the marmousi
velocity model. |
author2 |
Souza, Samuel Xavier de |
author_facet |
Souza, Samuel Xavier de Santana, Carla dos Santos |
format |
masterThesis |
author |
Santana, Carla dos Santos |
author_sort |
Santana, Carla dos Santos |
title |
Workload scheduling analysis in geophysical numerical methods |
title_short |
Workload scheduling analysis in geophysical numerical methods |
title_full |
Workload scheduling analysis in geophysical numerical methods |
title_fullStr |
Workload scheduling analysis in geophysical numerical methods |
title_full_unstemmed |
Workload scheduling analysis in geophysical numerical methods |
title_sort |
workload scheduling analysis in geophysical numerical methods |
publisher |
Universidade Federal do Rio Grande do Norte |
publishDate |
2020 |
url |
https://repositorio.ufrn.br/handle/123456789/30253 |
work_keys_str_mv |
AT santanacarladossantos workloadschedulinganalysisingeophysicalnumericalmethods |
_version_ |
1773963862281814016 |
spelling |
ri-123456789-302532020-10-11T07:39:01Z Workload scheduling analysis in geophysical numerical methods Santana, Carla dos Santos Souza, Samuel Xavier de Bianchini, Calebe de Paula Duarte, Ângelo Amâncio Chauris, Hervé Barros, Tiago Tavares Leite Escalonamento de tarefas FWI LSM Work-stealing Centralizado dinâmico Descentralizado estático Descentralizado dinâmico Computação de alto desempenho MPI The tasks organization among computational nodes affects the performance of the program. In computationally expensive applications such as geophysical problems, the impact is more significant. The imbalance caused by inefficient task scheduling can generate an application non-scalable. Therefore this work investigates the workload scheduling in geophysical methods. Three types of workload scheduling will be compared: centralized dynamic (CD), decentralized static (DS), and decentralized dynamic (DD). The CD was implemented with the technique master-slave, where the master node is responsible for the distribution of the tasks to other nodes called slaves. The DS was implemented with an equal division of tasks before the execution of them. The DD implements the work-stealing method proposed by Assis et al. (2019), where an idle node can steal the tasks of an overloaded node. The principal geophysical method used was 2D Full waveform inversion (FWI) with the acoustic wave. To analyze the performance of workload scheduling methods, we employ a synthetic velocity model and present speedup, efficiency, and load distribution plots generated with different model sizes and different quantity of nodes. The FWI code and the workload scheduling methods were implemented in C with distributed memory parallelization and using the message passing interface (MPI) library. With the results of the workload scheduling methods in FWI, it was applied the workstealing (because this technique presented the more effective performance) in another geophysical problem: Least-squares migration (LSM). We used the LSM with DS implemented by Chauris and Cocher (2017) to compare with the LSM with work-stealing. The LSM code and DS used in this problem were implemented in Fortran and the workstealing in C. The communication between the nodes was implemented using MPI. To analyze the performance of workload scheduling methods in LSM, we used the marmousi velocity model. A organização das tarefas entre os nós computacionais impacta diretamente no desempenho da aplicação. Considerando aplicações computacionalmente caras como os problemas geofísicos o impacto é mais significante. O desbalancemanto causado por um escalonamento ineficiente das tarefas pode tornar a aplicação não escalavel. Sendo assim esse trabalho investiga o escalonamento de tarefas em métodos geofísicos. Esse trabalho compara três tipos de escalonadores: centralizado dinâmico (CD), descentralizado estático (DS, do inglês Decentralized Static), descentralizado dinâmico (DD). O CD foi implementado através da técnica mestre-escravo o qual possui um nó mestre responsável pela distribuição das tarefas entre os outros nós chamados de escravos. O DS é a divisão igualitária da quantidade de tarefas entre os nós antes do início da execução delas. O DD utilizado implementa o método work-stealing proposto por Assis et al. (2019), qual o nó ocioso rouba a tarefa de um nó sobrecarregado. O principal método geofísico utilizado foi a inversão completa da forma de onda (FWI, do inglês Full Waveform Inversion) 2D com a onda acústica. Para analisar o desempenho dos escalonadores de tarefas foram gerados gráficos de speedup, eficiência e distribuição das tarefas entre os nós. O tamanho do problema foi definido a partir de modelos de velocidades sintéticos de diferentes tamanhos executados para diferentes quantidades de nós. O código do FWI e do escalonadores foram implementados em C em memória distribuída usando a biblioteca MPI. A partir dos resultados do FWI, foi aplicado o work-stealing (por ter apresentado comportamento mais eficiente) em outro problema geofisico: a migração por mínimos quadrados (LSM, do inglês Least Square Migration). Para comparar a implemetação LSM com work-stealing usamos o LSM com DS proposto por Chauris and Cocher (2017). O código do LSM e DS utilizado neste problema foram implementados em Fortran e o workstealing em C. A comunicação entre os nós foi implementado usando MPI. Para analisar a desempenho do escalonador em LSM, usamos modelo de velocidades marmousi. 2020-10-05T17:58:48Z 2020-10-05T17:58:48Z 2020-01-13 masterThesis SANTANA, Carla dos Santos. Workload scheduling analysis in geophysical numerical methods. 2020. 78f. Dissertação (Mestrado em Engenharia Elétrica e de Computação) - Centro de Tecnologia, Universidade Federal do Rio Grande do Norte, Natal, 2020. https://repositorio.ufrn.br/handle/123456789/30253 pt_BR Acesso Aberto application/pdf Universidade Federal do Rio Grande do Norte Brasil UFRN PROGRAMA DE PÓS-GRADUAÇÃO EM ENGENHARIA ELÉTRICA E DE COMPUTAÇÃO |