Meshing the Universe: Integrating Analysis in Cosmological Simulations

Mesh tessellations are indispensable tools for analyzing point data because they transform sparse discrete samples into dense continuous functions. Meshing the output of petascale simulations, however, can be as data-intensive as the simulations themselves and often must be executed in parallel on the same supercomputers in order to fit in memory. To date, however, no general-purpose large-scale parallel tessellation tools exist. We present a prototype method for computing such a Voronoi tessellation in situ during cosmological simulations. In principle, similar methods can be applied to other computational geometry problems such as Delaunay tetrahedralizations and convex hulls in other science domains. We demonstrate the utility of our approach as part of an in situ cosmology tools framework that runs various analysis tools at selected time steps, saves results to parallel storage, and includes visualization and further analysis in a widely used visualization package. In the example highlighted in this paper, connected components of Voronoi cells are interrogated to detect and characterize cosmological voids.