Powered by OpenAIRE graph
Found an issue? Give us feedback
image/svg+xml art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos Open Access logo, converted into svg, designed by PLoS. This version with transparent background. http://commons.wikimedia.org/wiki/File:Open_Access_logo_PLoS_white.svg art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos http://www.plos.org/ RUC. Repositorio da ...arrow_drop_down
image/svg+xml art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos Open Access logo, converted into svg, designed by PLoS. This version with transparent background. http://commons.wikimedia.org/wiki/File:Open_Access_logo_PLoS_white.svg art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos http://www.plos.org/
image/svg+xml art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos Open Access logo, converted into svg, designed by PLoS. This version with transparent background. http://commons.wikimedia.org/wiki/File:Open_Access_logo_PLoS_white.svg art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos http://www.plos.org/
image/svg+xml Jakob Voss, based on art designer at PLoS, modified by Wikipedia users Nina and Beao Closed Access logo, derived from PLoS Open Access logo. This version with transparent background. http://commons.wikimedia.org/wiki/File:Closed_Access_logo_transparent.svg Jakob Voss, based on art designer at PLoS, modified by Wikipedia users Nina and Beao
versions View all 4 versions
addClaim

Soft-fault recovery in MPI applications

Authors: Fernández Rey, David;

Soft-fault recovery in MPI applications

Abstract

[Resumo] Os sistemas actuais de computación de altas prestacións (HPC) están formados por miles de núcleos de procesadores, e espérase que este número aumente ata os millóns nun futuro cercano. Cun número tan elevado de procesadores, o tempo medio entre fallos (MTBF) pode chegar a reducirse tanto que a maioría de computacións científicas non terían tempo de completar a súa execución antes de que ocorrese un fallo. Polo tanto, é crítico o desenvolvemento de sistemas tolerantes e resilientes a fallos, para garantizar a finalización e integridade das aplicacións masivamente paralelas. O Grupo de Arquitectura de Computadores (GAC) da UDC propuxo unha solución (Controller/comPiler for Portable Checkpointing - CPPC) para convertir de xeito transparente aplicatcións MPI xenéricas en aplicacións tolerantes a fallos, basándose nun esquema de checkpointing e reinicio. CPPC foi posteriormente extendido por Nuria Losada en CPPC-resilience coa finalidade de crear aplicacións MPI resilientes, é dicir, aquelas que son capaces de detectar e reaccionar a fallos sen abortar a aplicación, de xeito que os procesos superviventes non necesitan ser reiniciados. Isto logrouse mediante un protocolo de logging de mensaxes e o uso dunha interfaz de tolerancia a fallos, ULFM (User Level Failure Mitigation), proposta para adición ao estándar MPI. Sen embargo, este sistema non xestiona os errores soft de maneira eficiente, xa que mata e reinicia os procesos fallados por completo cando non é necesario, xa que este tipo de errores teñen natureza transitoria. A meta deste TFG é extender e adaptar CPPC-resilience para poder manexar os errores soft eficientemente, sen ter que reiniciar os procesos fallados. Esta proposta foi evaluada utilizando 3 aplicacións MPI con diferentes características, conseguindo unha redución nos tempos de recuperación tras un erro soft de entre un 2 e un 44 por cento, dependendo do número total de procesos involucrados.

[Abstract] Current high-performance computing (HPC) systems are comprised of thousands of CPU cores, and this number is expected to grow into the millions in the near future. With such an elevated number of processors, the mean time between failures (MTBF) can become so small that most scientific applications will not have time to complete their execution before a failure occurs. It is therefore critical to develop fault tolerance and resilience mechanisms in order to guarantee the completion and integrity of massively parallel applications. University of A Coruña’s Computer Architecture Group (GAC) proposed a solution (Controller/comPiler for Portable Checkpointing - CPPC) in order to transparently convert generic MPI applications into fault tolerant applications, based on a checkpoint-restart scheme. CPPC was extended by Nuria Losada into CPPC-resilience in order to make resilient MPI applications, that is, those that are capable of detecting and reacting to failures without aborting the application, such that survivor processes don’t have to be restarted. This was accomplished by means of a logging protocol and the usage of a proposed fault tolerance interface addition to the MPI standard (User Level Failure Mitigation). However, this system cannot handle soft errors efficiently, since it kills and respawns the failed processes entirely when it is not necessary, as these errors are transient in nature. The object of this project is to extend and adapt CPPCresilience in order to handle soft errors in a more efficient manner, without having to respawn the failed processes. This proposal has been evaluated using 3 MPI applications with different characteristics, achieving a decrease in recovery times after a soft error ranging from 2 to 44 percent, depending on the total number of processes involved.

Country
Spain
Related Organizations
Keywords

Resilience, Soft errors, Tolerancia a los fallos, Fault tolerance, Parallelism, CPPC, Resiliencia, MPI, Erros soft, Computación de altas prestacións, High-performance computing, Paralelismo, ULFM

  • BIP!
    Impact byBIP!
    selected citations
    These citations are derived from selected sources.
    This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
    0
    popularity
    This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network.
    Average
    influence
    This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
    Average
    impulse
    This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network.
    Average
Powered by OpenAIRE graph
Found an issue? Give us feedback
selected citations
These citations are derived from selected sources.
This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
BIP!Citations provided by BIP!
popularity
This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network.
BIP!Popularity provided by BIP!
influence
This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
BIP!Influence provided by BIP!
impulse
This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network.
BIP!Impulse provided by BIP!
0
Average
Average
Average
Green