University of California, Irvine
HIPerWall: New Vistas in Scientific Visualization
Using satellite imagery taken at a resolution of one meter per pixel, the HIPerWall displays a 25.60 x 8.00 kilometer (15.91 x 4.97 mile) region of the Earth’s surface.
The HIPerWall is one component of a multi-geographic, high-performance computing environment consisting of many computational clusters, each of which performs a unique job.
In this architectural model, the enormous computational capacity of the HIPerWall cluster is reserved solely for the specialized functions appropriate to a display subsystem. The other cluster resources, dedicated to running additional scientific applications, are joined together by the OptIPuter optical network to create a virtual supercomputer of staggering power.
To date, UCI funding has allowed researchers to display two brain imaging projects and a comparison of geoscience climate models on the HIPerWall. Much of the work thus far has involved getting the data in an appropriate form on the display and learning to interact with data on a much larger scale than has previously been done.
Some of the research projects that will run on HIPerWall over the coming years will be conducted by UCI researchers. Others will be research projects based at other academic institutions, the national labs, and businesses.
Kuester indicates the following types of research projects make a good fit for the visualization capabilities of the HIPerWall and are anticipated for the near future:
HIPerWall team members, from left, Christopher Knox, Falko Kuester, and Frank Wessel view the microscopic structure of the brain of a rat.
- Biomedical data exploration (brain imaging). Brain scans are typically viewed on photographic film or desktop monitors as a small group of images, each of which represents one cross-sectional slice of the brain. This is problematic because the person evaluating the images can see only six to 10 slices at a time. The HIPerWall has the resolution to display all of the images from a brain scan simultaneously, thus retaining both the context and the details of the entire scan sequence. Even more important is the ability to compare brain images between different patients. The HIPerWall has the resolution needed to display multiple slices from many different patients to help medical specialists compare and correlate the characteristics of the feature being studied. And instead of having an intern running to the lightboard to exchange one set of films for another, a simple mouse-click is all it takes to step through successive sets of images on the HIPerWall.
- Visualization of earth system science simulations. The human eye is much better than strict numerical analysis at catching anomalies and trends when viewing the simultaneous display of sequences from several different climate models. Very large displays like the HIPerWall will help geoscience researchers identify interesting characteristics for more detailed study.
- Remote reconnaissance and disaster assessment. Using detailed before-and-after satellite and aerial photography images, rescue workers and recovery planners can quickly assess damage, determine passable access routes, and understand the magnitude of the disaster. The post-Katrina images Jenks and his colleagues built from NOAA photos to display on the HIPerWall were manually tiled using Adobe Photoshop, but developing an integrated workflow will streamline the process of getting imagery up on the HIPerWall, facilitating quicker response in future emergencies.
- Interactive three-dimensional terrain visualization. Familiar applications like Google Earth and NASA World Wind demonstrate how valuable 3D terrain visualization is even at fairly low resolutions. Extending the concept to very high resolutions involves dealing with correspondingly large terrain data files, so the challenge is to render the appropriate views seamlessly across the tiles. Out-of-core rendering with varying levels of detail makes it possible to fetch only the needed portions of the terrain dataset and render the appropriate mesh complexity for the current view and zoom level.
A New Collaborative Workflow
The HIPerWall represents an innovation in the way that scientists can collaborate seeing visual images in grand scale. It’s infrastructure and middleware are all network-based to allow easy export and import of imagery and data. This will allow teleconference-style meetings with much greater data visualization capability than current approaches.
“Now that HIPerWall is fully functional, our highest priority is bringing in collaborators from academia, from industry, and from the national labs, to apply this research instrument to their respective problem domains,” Kuester says.