Science Systems and Applications, Inc. - Science and Technology with Passion

High-performance computing

SSAI employees support the high performance computing capabilities of the Atmospheric Science Data Center at NASA Langley Research Center. Credit: NASA LaRCSSAI employees support the high performance
computing capabilities of the Atmospheric Science
Data Center at NASA Langley Research Center.
Credit: NASA LaRC
SSAI’s customers collect large amounts of critical data—data that help scientists to unravel the mysteries of the world around us. SSAI is dedicated to protecting and maintaining the integrity of these data with high-quality, customer-focused data stewardship. We process, archive, and distribute large quantities of data quickly and efficiently. In some areas, we are responsible for managing over 3 petabytes of data. We have the expertise in high-performance computing to provide reliable data management and innovative solutions and methods that meet or exceed customer needs.

SSAI’s Global Modeling and Assimilation Office (GMAO) Software Infrastructure Team improved the performance of the GEOS-5 Global Circulation Model (GCM) in various ways over the past year. Recent successes include modifying the GCM to in order to utilize PGI CUDA Fortran with NVIDIA graphics processing units (GPUs) to speed up execution of selected expensive physics kernels, rendering the entire physics calculation process approximately two times faster. We also nearly doubled the speed for computational times in solar radiation code; moreover, this code has been ported as a test case for new accelerators, showing promise of even greater gains.

At NASA Langley’s Atmospheric Science Data Center (ASDC), SSAI employees compute and process over 1.4 terabytes of data each day. Data flow in from various Earth-observing instruments, such as the Clouds and Earth’s Radiant Energy (CERES), and we make sure that all of the data are ingested securely and rapidly.

A new Input/Output (I/O) server paradigm was implemented in the Message Passing Interface for the GCM in order to speed up expensive model output writing at very high resolution. Before the introduction of the I/O server, writing of the model output was 50-60% of model run time; now, it occurs simultaneously as the model runs on separate nodes, effectively doubling the wall-clock performance.