Dynamic vs static implementations of MPI on new trend computer architectures

Hyacinthe Nzigou Mamadou, Takeshi Nanri, Kazuaki Murakami

    Research output: Chapter in Book/Report/Conference proceedingConference contribution

    Abstract

    Nowadays, multi-core chips are becoming the norm for the computer market. When one considers MPI collective communications with several processor cores on a node, one also has to consider the impact of such new architecture on widely used MPI implementations, in order to realize the communication efficiently and achieve high performance computing. This paper provides a succinct comparative analysis between dynamic and static MPI implementations on multi-core computers. MPI Alltoall collective operation is used as an example to handle the communication process involved in a load unbalanced case, and then among different communicators. The experiments performed on clusters equipped with high performance interconnects, such as Infiniband and Gigabit Ethernet, revealed that the dynamic version is more likely to be adapted with the general and random behaviors of the system for this new trend of computer architectures.

    Original languageEnglish
    Title of host publicationWMSCI 2008 - The 12th World Multi-Conference on Systemics, Cybernetics and Informatics, Jointly with the 14th International Conference on Information Systems Analysis and Synthesis, ISAS 2008 - Proc.
    Pages113-118
    Number of pages6
    Publication statusPublished - Dec 1 2008
    Event12th World Multi-Conference on Systemics, Cybernetics and Informatics, WMSCI 2008, Jointly with the 14th International Conference on Information Systems Analysis and Synthesis, ISAS 2008 - Orlando, FL, United States
    Duration: Jun 29 2008Jul 2 2008

    Publication series

    NameWMSCI 2008 - The 12th World Multi-Conference on Systemics, Cybernetics and Informatics, Jointly with the 14th International Conference on Information Systems Analysis and Synthesis, ISAS 2008 - Proc.
    Volume5

    Other

    Other12th World Multi-Conference on Systemics, Cybernetics and Informatics, WMSCI 2008, Jointly with the 14th International Conference on Information Systems Analysis and Synthesis, ISAS 2008
    CountryUnited States
    CityOrlando, FL
    Period6/29/087/2/08

    All Science Journal Classification (ASJC) codes

    • Artificial Intelligence
    • Computer Networks and Communications

    Fingerprint Dive into the research topics of 'Dynamic vs static implementations of MPI on new trend computer architectures'. Together they form a unique fingerprint.

    Cite this