Interactive Weather Simulation and Visualization on a Display Wall with Many-Core Compute Nodes

The task of Weather forecasting requires visualization of large data volumes. Numerical Weather Prediction models (NWP) are essential tools, and are typically run in batch systems for predetermined areas where data is available only at the end of each complete run . The NWPs are often run for as large an area as the available computer resources allow, this limits the resolution and area covered. We report on a system using a high-resolution, wall-sized tiled display where a user can select a region of interest, and have an NWP model run on-demand for the selected area. The NWP is only run for the requested areas, and may therefore also be run with high spatial resolution. Visualizing these large datasets using a display wall enables the user to view high level of detail for a larger area than regular displays. To enable interactive use of NWP, we make extensive use of several many-core systems to drive the on-demand computational needs. We identify and document the bottlenecks and computational challenges the combination of interactivity and traditional batch oriented computing creates. The main bottleneck is identified as the execution time of the NWP and preparing data for visualization.