Heterogeneous Architectures for Earth Science Computing

Overview

The NASA Center for Climate Simulation (NCCS) at Goddard Space Flight Center supplies high-performance computing (HPC) and data services engineered to support climate simulations within NASA's Science Mission Directorate (SMD). Over the past 5 years, NCCS has deployed a scalable cluster approach in order to meet the application requirements of large-scale climate simulations. In addition, NCCS has continued its practice of exploring and experimenting with emerging technologies that could enhance its services for the climate community. Technologies of note include graphics processing units (GPUs) for high-resolution simulations, Hadoop file systems, MapReduce data analysis, and virtual environments applied to HPC.

Project Details

The NCCS Discover cluster at NASA Goddard currently features:

• 31,434 Intel processor cores
• 400 teraflops peak performance
• A mix of 20-gigabit-per-second and 40-gigabit-per-second InfiniBand interconnect technology
• A recent addition of 64 NVIDIA Tesla M2070 GPUs (28,672 total cores and 36.4 teraflops peak performance)

In addition to expanding its computational capabilities, NCCS has begun to explore the use of the Apache Software Foundation's Hadoop file systems and MapReduction operations for large-scale climate data analysis. The overarching goals of this project are: 1) to provide a preliminary qualitative and quantitative assessment of MapReduce as a means of enabling server-side, distributed analytics, and 2) to assess the relative merits of object storage versus high-performance block storage in meeting the large data management needs of the climate research community.

Results and Impact

NCCS has enabled multiple climate research success stories. Some examples include:

• High-fidelity, 3.5-kilometer, global simulations of cloud and hurricane predictions
• A comprehensive reanalysis of the last 30 years of weather/climate data as part of the Modern Era Retrospective-analysis for Research and Applications (MERRA) project
• Multi-century climate analysis for the Intergovernmental Panel on Climate Change

Role of High-End Computing Resources

The NCCS Discover cluster supports modeling and data assimilation efforts throughout SMD. NCCS has tailored its services to the specialized needs of climate applications, and has designed Discover cluster upgrades to support research in running high-fidelity, global climate models at unprecedented resolutions.