Português English
Contato
Publicado em: 22/09/2009

Defesa de Proposta de Tese em Processamento Paralelo e Distribuído dia 28/09 de Márcia Cristina Cera

UNIVERSIDADE FEDERAL DO RIO GRANDE DO SUL
INSTITUTO DE INFORMÁTICA
PROGRAMA DE PÓS-GRADUAÇÃO EM COMPUTAÇÃO


DEFESA DE PROPOSTA DE TESE


Aluno: Márcia Cristina Cera
Orientador: Prof. Dr. Philippe Olivier Alexandre Navaux
Titulo: Supporting Malleability in Current Parallel Architectures.
Linha de Pesquisa: Processamento Paralelo e Distribuído

Data: 28/09/2009

Hora: 10hs

Local: Auditório José Mauro Volkmer de Castilho – Instituto de Informática – UFRGS

Banca Examinadora:
Prof. Dr. Jairo Panetta (CPTEC/INPE)
Prof. Dr. Gerson Geraldo Homrich Cavalheiro (UFPel)
Prof. Dr. Alexandre da Silva Carissimi (UFRGS)

Presidente da Banca: Prof. Dr. Philippe Olivier Alexandre Navaux

Resumo:
Malleability adds some flexibility to parallel applications, so that malleable applications can adapt themselves according to variations in the amount of resources available. Providing such applications upon current parallel architectures became possible thanks to evolutions of the runtime system and the programming environment. The handling of dynamic resources has been included in the latest versions of some resource management systems as well as some flexibility to jobs allocation. Programming environments also included some flexibility enabling the development of growing and shrinking operations, providing an on-the-fly adaptation according to resource availability. Our work focus on MPI malleable applications, named dynamic MPI applications, executed upon the current parallel architectures thanks to OAR resource manager. Dynamic MPI applications perform growing operations spawning dynamic MPI processes to load new resources. In shrinking, the processes running on resources announced as unavailable are safe stopped through a fault tolerance procedure. OAR informs the applications about the current resources’ availability. It identifies the unused resources and make them available to execute malleable jobs, aiming at improvements in resources utilization. This thesis proposal deals with two main goals:

(i) provide a framework for developing dynamic MPI applications, and

(ii) execute such applications upon current parallel architectures thanks to the support given by the OAR resource manager. The contribution of this work is twofold: (i) expose challenges and requirements to develop malleable applications using features of the current MPI distributions, and (ii) reach improvements in resources utilization of the current parallel architectures through the execution of malleable jobs upon their unused resources.

Palavras-Chave: Malleability, Dynamic Resources, MPI