• WORKSHOP

    19 y 20 de Septiembre, Bogota D.C, Colombia
    La serie de talleres sobre programación paralela y de alto desempeño, ofrecidos por la universidad Distrital Francisco José de Caldas tienen como objetivo proporcionar una introducción completa y práctica a técnicas y tecnologías de programación y optimización en computación paralela basadas en estándares y marcos abiertos con el fin de utilizar totalmente las capacidades de escalamiento…

    DETALLES DEL EVENTO E INSCRIPCIONES
  • AWS CLOUD SERVICES FOR RESEARCH COMPUTING AND HPC

    Researchers are increasingly bringing their HPC workloads to the AWS cloud. In this presentation, we will visit AWS capabilities, real-world examples, and best practices, to shine a light on:
    - AWS services that make scientific computing accessible to everyone – from new HPC virtual server types to personal HPC clusters in the cloud;
    - New approaches to HPC, including serverless computing;
    - How AWS works with the research community and supports science.

    DETALLES DEL EVENTO E INSCRIPCIONES
  • SEMINARIO

    Seminario sobre optimización de aplicaciones MATLAB dentro del marco del HPC UD'17.

    DETALLES DEL EVENTO E INSCRIPCIONES
esenfrpt
Lunes, 13 Marzo 2017 09:17

Efficiently Energetic Acceleration EEA-Aware For Scientific Aplications of Large-Scale On Heterogeneous Architectures

Escrito por
Valora este artículo
(0 votos)

Abstract:

Heterogeneous parallel programming has two main problems on large computation systems: one is the increase of power consumption on supercomputers in proportion to the amount of computational resources used to obtain high performance, and the second problem is the underuse these resources by scientific applications with improper distribution of tasks. Select the optimal computational resources and make a good mapping of task granularity is the fundamental challenge for build the next generation of Exascale Systems. This research propose an integrated energy-aware scheme called efficiently energetic acceleration (EEA) for large-scale scientific applications running on heterogeneous architectures.

The EEA scheme uses statistical techniques to get GPU power levels to create a GPU power cost function and obtains the computational resource set that maximizes energy efficiency for a provided workload. The programmer or load balancing framework can use the computational resources obtained to schedule the map parallel task granularity in a static time. This poster proposes an integrated energy-aware scheme called Efficiently Energetic Acceleration (EEA) for scientific applications of large-scale on heterogeneous architectures. The EEA structure has a workflow of three steps as shown in Figure 2. In the first step, the data is captured in runtime and executed the enerGyPU or enerGPhi monitor tool in parallel with the scientific application using different combinations of computational resources, applying the power capping technique for nodes with multi-GPUs. In the second step, the data visualization and statistics characterization used a separate level of enerGyPU and enerGPhi monitor tool for analysis the key factors by results of each experiment in terms of energy efficiency and estimated the power levels. A deep description of monitor structure and utilization is present by García John et al. in [2], [3]. Finally using the data collected by monitors cost functions are build and model prediction system running for obtaining the optimal computational resources in a static time to mapping parallel task granularity of scientific applications on heterogeneous architectures. 1,2 Esta dirección de correo electrónico está siendo protegida contra los robots de spam. Necesita tener JavaScript habilitado para poder verlo., 1,3 ejhernandezb @udistrital.edu.co, 2 Esta dirección de correo electrónico está siendo protegida contra los robots de spam. Necesita tener JavaScript habilitado para poder verlo., 1 Esta dirección de correo electrónico está siendo protegida contra los robots de spam. Necesita tener JavaScript habilitado para poder verlo.

 

Visto 209 veces Modificado por última vez en Lunes, 20 Marzo 2017 09:46

Deja un comentario

Asegúrese de introducir toda la información requerida, indicada por un asterisco (*). No se permite código HTML.

APOYAN WORKSHOP "HPC UD ‘17”

  • 1A
  • 2A
  • 3A
  • 4A
  • 5A
  • 6A
  • 7A
  • 8A