Presentation
Performance and Communication Modeling for Exascale Proxy Architecture in Aspen
Author/Presenter
Event Type
Workshop
W
Architectures
Collaborative Environments
Parallel Programming Languages, Libraries, and Models
Simulation
Workflows
TimeMonday, November 12th9:20am - 9:40am
LocationD173
DescriptionThe performance capability of current high-performance computing cores are already at their threshold, leading to an unprecedented increase in number and complexity of exascale computational processors. Anticipating the performance and communication modeling required for millions of exascale cores necessitates profiling and modeling of proxy exascale applications on current supercomputers to help us pin-point to the application's hardware requirements to perform corresponding what-if analysis. This has to be done while considering both applications and hardware, leading to efficient co-design for exascale proxy architectures.
In this paper, we study and profile the computation and communication of one of the exascale proxy applications -- Co-design for Molecular Dynamics (CoMD). We model one of the proposed exascale homogeneous architecture in the Aspen machine modeling language and develop the model into a refined proxy architecture. We model the CoMD's computational and communication requirements on Exascale proxy architecture in the Aspen application modeling language to study the scalability analysis. We also present an analysis about communication profiling for CoMD in Aspen, which guides the efficient communication mapping of the application on the proxy architecture.
In this paper, we study and profile the computation and communication of one of the exascale proxy applications -- Co-design for Molecular Dynamics (CoMD). We model one of the proposed exascale homogeneous architecture in the Aspen machine modeling language and develop the model into a refined proxy architecture. We model the CoMD's computational and communication requirements on Exascale proxy architecture in the Aspen application modeling language to study the scalability analysis. We also present an analysis about communication profiling for CoMD in Aspen, which guides the efficient communication mapping of the application on the proxy architecture.
Archive
Author/Presenter
