| Beowulf clusters and Grid computing have commoditized computing and one of the largest users of this commodity simulate large physical and biological systems involving billions of entities. Success of such ultrascale simulations depends essentially on the visualization and analysis of massive multivariate data sets. However with increasing system size, researchers are being overwhelmed with data and information. Scientific visualization and data mining are two powerful tools that can help a researcher explore and understand their data. However, exploratory and hence interactive visualization and mining of billion entity datasets pose enormous computational challenges. To address this issue, we are have developed scalable and parallel scientific visualization and data mining algorithms, to aid computational scientists performing billion-particle simulations of materials. We introduce novel algorithms that use hierarchical data abstraction for data culling, probabilities for occlusion culling, multiple levels-of-detail, and parallel and distributed variations of our techniques. These algorithms have been successfully implemented in a scientific visualization application that has been disseminated into the community and to visualize some of the largest systems in the world. We also present graph based clustering and feature detection algorithms to identify and track topological and structural anomalies of chemical bond networks in materials. |
