Visualize This: Lessons from the Front-lines of High Performance Visualization

This paper presents a comprehensive workflow to address two major factors in multivariate multidimensional (MVMD) scientific visualization: the scalability of rendering and the scalability of representation (for perception). Our workflow integrates the metrics of scientific computing and visualization across di fferent STEM domains to deliver perceivable visualizations that meet scientists’ expectations. Our approach attempts to balance the performance of MVMD visualizations using techniques such as sub-sampling, domain decomposition, and parallel rendering. When mapping data to visual form we considered: the nature of the data (dimensionality, type, and distribution), the computing power (serial or parallel), and the rendering power (rendering mechanism, format, and display spectrum). We used HPC clusters to perform remote parallel processing and visualization of large-scale data sets such as 3D point clouds, galaxy catalogs, and airflow simulations. Our workflow brings these considerations into a structured form to guide the decisions of visualization designers who deal with large heterogeneous data sets.