Human Space Flight

Overview

Prior missions to Mars have been limited to robotic payloads of about one metric ton, however payloads of 20–30 metric tons will be required for future human exploration of the Martian surface. Conventional deceleration techniques using parachutes cannot be applied to reduce speed and land vehicles of this scale. Instead, novel retropropulsion concepts—where the engines are fired in the direction of travel—must be used to safely decelerate these larger large vehicles as they descend through the thin Martian atmosphere. NASA engineers are exploring the complex aero-propulsive effects of retropulsion approaches with a goal to ultimately understand guidance and control requirements for these future missions.

a picture of a retropropulsion visualization example produced with OSPRay

We perform simulations on the Summit supercomputer at Oak Ridge National Laboratory with NASA’s FUN3D software to efficiently characterize the effects of throttle settings, grid resolution, freestream velocity, and reacting-gas chemistry modeling. In order to visualize, analyze, and archive these data, we transfer solution datasets as large as one petabyte across the country to the NASA Advanced Supercomputing (NAS) facility at NASA’s Ames Research Center. The data size, and the approach of writing the data in FUN3D’s native domain decomposition on Summit, make the visualization phase particularly challenging.

Project Details

In our visualization approach, we adapted the workflow to accommodate the FUN3D output files written by each subdomain process in the Summit simulations, with each run utilizing between 5,532 and 15.912 subdomains. Post-processing is done on the hyperwall visualization cluster, part of the Pleiades system at NAS, and where practical, processing of each of the subdomains is done independently to exploit the opportunities for parallelism in preparation for rendering.

In order to visualize, analyze, and archive these data, we transfer solution datasets as large as one petabyte across the country to the NASA Advanced Supercomputing (NAS) facility at NASA’s Ames Research Center.

Patrick Moran, NASA Ames Research Center

We take two visualization rendering approaches: one CPU based and one GPU based. For the CPU approach, we use the Open-Source library Intel OSPRay, which does high-quality ray-traced rendering, which can render at interactive rates for many scenes. The contributions from all the subdomains (e.g., parts of a cutting plane) are combined into a single, full-domain input to our renderer. We are investigating use of the MPI distributed rendering mode of OSPRay to generate images in a more parallel, scalable manner.

For the GPU-based rendering, we use the NVIDIA IndeX 3D Volumetric Visualization Framework, which is a distributed renderer. For the high-resolution post-processing runs, we use 40 nodes of the NAS hyperwall system, where each node has an NVIDIA Quadro RTX 6000 GPU. As part of an ongoing collaboration with NASA, NVIDIA developed an IndeX reader plugin capable of reading FUN3D native subdomain solution files. Once the data are loaded, we volume render scenes at interactive rates, even for the high-resolution runs where the computational mesh has several billion unstructured elements.

Results and Impact

The high-end visualization approaches being developed and applied in this effort are critical to understanding the complex flow field phenomena associated with reacting-gas retropropulsion for human-scale Mars landers. The ability to visually analyze and interrogate the intricate time-dependent interplay among boundary layer mechanisms, reacting-gas physics, multiple shear layers, turbulence phenomena, shock structures, and other features will be critical to reach an understanding of the engineering requirements ultimately needed to develop large databases of sufficient fidelity to support the detailed design of such a vehicle.

Why HPC Matters

a picture of a retropropulsion visualization example produced with IndeX

Due to sheer size and complexity, data generated on a supercomputer often also requires supercomputing resource for the visualization and data analysis. We are employing state-of-the-art CPU- and GPU-based technologies to provide timely, high-quality visualization and post-processing support of multiple petabytes of data. This work simply would not be possible without the type of networking, high-performance file systems, and processing capabilities provided by the supercomputing resources at NAS.

What's Next

In our future work, we will continue to improve the distributed rendering capabilities in order to make our analysis more interactive and informative. We also aim to make these tools easier to use directly by the computational scientists who are producing these large datasets.

More Information

Researchers

  • Patrick Moran, NASA Ames Research Center
  • Gabriel Nastac, NASA Langley Research Center
  • Paul Kolano, NASA Ames Research Center
  • Eric Nielsen, NASA Langley Research Center

Acknowledgements

The team thanks Intel for producing OSPRay, and for making the library available as an Open-Source project, as well as the NVIDIA Advanced Rendering Center for IndeX, their development of the FUN3D extension, and their ongoing collaboration. Resources used for this work were provided by the NASA Advanced Supercomputing Division at NASA Ames Research Center and the Oak Ridge Leadership Computing Facility at the Oak Ridge National Laboratory, which is supported by the Office of Science of the U.S. Department of Energy under Contract No. DE-AC05-00OR22725.