large data visualization initiative

largedatavisualizationinitiative
nsf-99 ©

p r o j e c t s u m m a r y

Advances in computing technology are making simulation as important to science today as theory and experiment have been in the past. Unfortunately this success is overwhelming the scientific community. The amount of data that can be acquired or generated by these advancing technologies is enormous. Scientists are drowning in their data. While scanning and raw computing technology provide the means to acquire or generate sequences of terabyte datasets, other technologies, e.g. storage, networking and graphics, have not kept up, creating technological bottlenecks. Because of these bottlenecks, more data is produced than can be analyzed or visualized. Valuable information is being lost or neglected because of these deficiencies. Additionally, scientists are painfully aware of this technological bottleneck, and it affects how they conduct their science. It slows the rate of data acquisition, therefore scanning devices are not being used to their full capacity. and calculations that could generate higher-dimensional data, e.g. vector and tensor, are not being performed.

The goal of this project is to address the visualization bottleneck by developing multiresolution software tools which will facilitate understanding of large-scale experimental and simulation datasets; thus providing advances in fluid dynamics and neuroscience. This will be accomplished by

interacting with scientists in the fields of fluid dynamics and neuroscience in order to better understand their working domain and visualization needs,
developing new data analysis and visualization algorithms for large-scale datasets,
producing robust implementations of these algorithms,
training the scientists to utilize the tools and working with the scientists to improve them.

The technical approach of the project is based on the understanding that the graphics workstations available to most scientists are incapable of visualizing giga- and tera-scale datasets. It is also clear that for the foreseeable future our ability to generate data will outpace our ability to render and visualize it.

Therefore it is essential to develop methods for representing, segmenting and compressing these enormous datasets into a form that can be processed on today's graphics computers. Multiresolution representations and algorithms are at the core of the methods and tools to be developed. A multiresolution approach ensures that visualization techniques will scale to effectively cope with today's enormous datasets. Multiresolution methods accomplish this by focusing resources in those regions of the dataset most important to the user. Multiresolution modeling and display techniques will allow a user to view a single dataset at a coarse level-of-detail, and easily provide pertinent details only in those regions of greatest interest. During analysis, multiresolution methods can allocate more computational resources in segments of the data with the greatest rates of change or containing specific properties.

This project will produce a set of multiresolution software tools for processing and visualizing large-scale N-dimensional (scalar, vector and tensor) volumetric datasets (NDVDs), as well as large-scale triangle meshes. Specifically, tools for segmenting and interpolating NDVDs using level-set methods, extracting semi-regular meshes directly from volume datasets, compressing triangle meshes, and volume rendering NDVDs will be developed. The integration of these tools will provide interactive capabilities to scientists that will allow them to more effectively analyze and visualize their large-scale datasets.

As software tools become available they will be transferred to scientists at Caltech's Center for Simulation of Dynamic Response of Materials, which is part of the DOE's Accelerated Strategic Computing Initiative, and to scientists at the National Center for Microscopy and Imaging Research (UC San Diego) and Caltech's Biological Imaging Center, which are both participating in the NPACI Neuroscience Initiative. The scientists will be trained to use the tools, and will provide feedback towards improving the tools and producing additional enhancements to them.