Return to the Table of ContentsObjective: To develop techniques which permit parallel application programs to efficiently generate visual output at runtime.
Approach: The traditional approach to visualizing results from supercomputer-based computations has been to dump the output data to files for subsequent post-processing and display on a graphics workstation. For memory-intensive or time-dependent computations, this can result in massive volumes of data which must be moved across the network. We are developing algorithms and methodologies which exploit the available parallelism to generate the graphics in place. We are particularly interested in techniques which can be inserted directly into parallel applications in the form of library calls. This approach offers live visual feedback for debugging, execution monitoring, analysis, and interactive steering. By using compressed image streams, we also reduce the amount of data which must be transferred across the network.
Accomplishment: We are incorporating our research results into a parallel graphics library known as PGL. PGL is targeted primarily toward distributed memory message passing architectures. A preliminary version is currently running on the Intel family of parallel machines. We have used PGL in conjunction with a parallel Direct Simulation Monte Carlo (DSMC) code to provide runtime visualization of the rarefied gas plume emanating from a pair of reaction-control jets being fired in a vacuum. The accompanying videotape shows the results of the first 200 timesteps in the simulation. The particles have been color-coded according to their velocity magnitudes. The simulation and visual results were produced on Langley's Intel Paragon using 32 processors.
Significance: This is the first demonstration of PGL in conjunction with a parallel scientific application. The immediate visual feedback provided by PGL will aid in the development and debugging of more sophisticated versions of the parallel DSMC code, which will be used to simulate a variety of problems involving rarefied flows. PGL can also provide a visual record of simulation runs without having to save the intermediate results from every timestep. For the example described here, the animation file is an order of magnitude smaller than the original data.
Status/Plans: Development of PGL continues, and we plan to release it to the HPCCP community when it becomes sufficiently robust. Ongoing research efforts are focused on improving the efficiency and scalability of the underlying algorithms.
Points of Contact:
Richard G. Wilmoth
NASA Langley Research Center
(804) 864-4368
r.g.wilmoth@larc.nasa.gov
Return to the Table of Contents