Visualization

Today, visualization is mostly done as a separate step after simulation. A common practice is to down-sample simulation results to fit into the memory of a graphics workstation for visualization processing. Unfortunately, this approach requires massive amounts of simulation data to be moved between the simulation processes and the visualization process.  Therefore, we propose a tightly integrated and parallelized visualization process that runs on the exascale system itself.

© NASA Artist's view

This in-situ visualization works around the data transfer bottleneck, by operating on-the-fly on the simulation data while it is still in memory. A first challenge consists of coordinating visualization and simulation computations operating on the same data structures. The second challenge is to develop scalable and robust visualization algorithms.