Scalable Data Servers for Large Multivariate
Volume Visualization

Volumetric datasets with multiple variables on each voxel over multiple timesteps are often complex, especially when considering the exponentially large attribute space formed by the variables, and spatial and temporal dimensions. It is intuitive, practical, and thus often desirable, to interactively select a subset of the data from within that high-dimensional value space for efficient visualization. This approach is straightforward to implement if the dataset is small enough to be stored entirely in-core. However, to handle datasets sized at hundreds of gigabytes and beyond, this simplistic approach becomes infeasible. More sophisticated solutions are needed. In this work, we developed a system that supports efficient visualization of an arbitrary subset, selected by range-queries, of a large multivariate time-varying dataset. By employing specialized data structure and schemes of data distribution, our system can leverage a large number of networked computers as parallel data servers and guarantees a near optimal load-balance. We demonstrate our system of scalable data servers using a large dataset from a supernova simulation. This work has been published in IEEE Transactions on Visualization and Computer Graphics, Vol. 12, No. 5, 2006. pdf

In the above photograph of the Everest visualization facility at the Oak Ridge National Laboratory, we demonstrate an on-demand visualization environment for hypotheses-driven scientific research, supported by an installation of our scalable data servers.

Credits:
The TSI data was provided by John Blondin and Anthony Mezzacappa under the auspices of the DOE SciDAC Terascale Supernova Initiative.