Purdue project aims to move visual mountains, and quickly

  • December 8, 2008
  • Science Highlights

In an age of scientific visualization employing huge datasets, and of networked instruments that produce data in torrents, bandwidth is an issue. More of it may be available than ever before and it may be faster than in the past, but the pipeline remains relatively limited and expensive.

“Networking is not cheap,” said Purdue computer science Professor Chris Hoffmann. “Bandwidth is not very big at the moment. We’re working on technology to make all that, shall we say, affordable, particularly the networking.”

A system being developed at Purdue could make moving large, high-resolution scientific visualizations long distance over a network like the Internet manageable and it may open up new opportunities in interactive video conferencing and online collaboration as well, said Hoffmann, who also is director of Purdue’s Rosen Center for Advanced Computing, the research and discovery arm of Information Technology at Purdue (ITaP), the university’s central information technology organization.

The idea is to do the heavy lifting for a visualization on computers close to a big dataset being drawn upon, pack the result for shipping over a network, with no loss of fidelity, and send it to the user’s end. There, it gets unpacked locally, ready to run with fluidity comparable to streaming video—or television—but less consumption of bandwidth along the way.

The “lossless” technique uses features of the data to be visualized, and an understanding of them built into the algorithms behind it, to eliminate redundancy and create a compact single-image package sufficient for reconstructing a series of images on the receiving end, said Hoffmann’s colleague Voicu Popescu, also a computer science professor and, like Hoffmann, a member of Purdue’s Computer Graphics and Visualization Lab.

More and more, Hoffmann said, scientists want to look at all, or a big part, of a large dataset—data covering genomes, climate change factors, the interactions of elementary particles in physics, the stars in the heavens and more—to glean meaning and ultimately insight in a process he likened to a pyramid.

At the pointy peak, scientists examine large regions of the data at low resolution looking for areas of interest, delve further into them in the middle and discard some false positives while formulating hypotheses and theories, then look at surviving regions of interest at high resolution—the imposing base of the pyramid.

But the pyramid’s base and even the other stages are generally too unwieldy to move long distance via a network, Hoffmann said. So the idea is to cull the data at the source and transmit only the visualization output, images.

Popescu said that’s still no small matter. High-resolution scientific visualizations, and the other kinds of materials the researchers have in mind, can be weighty end products. They’re no YouTube videos. “You can reach very high resolutions,” Popescu said. “The images themselves are non-negligible size, which, multiplied by the frame rate, yields great bandwidth demands.”

Popescu used a movie camera analogy in describing the capabilities of the Purdue system, calling the compressed package “a super frame which would be sufficient to recreate a sequence of frames.” He said the aim is to achieve a highly interactive refresh rate of 60 frames per second or more.

The “camera” in this case is based on a non-pinhole model, he said. Instead of conventionally capturing the data for one frame at a time through a single virtual pinhole, the system records all the data for a visualization segment, that is several frames worth, en masse.

The system takes advantage of the computing power offered by today’s powerful, readily available and largely computer game-driven graphics processing unit chips, or GPUs, to do that with dispatch, and likewise to quickly unpack and display a visualization on the receiving end.

The goal is to make the system responsive enough for interactivity.

Besides visualization, it also might be used for online classes, conferencing and collaboration, allowing a professor, say, at Purdue’s West Lafayette campus to interact more naturally with students at the University’s Calumet campus, in Muncie, Bloomington or elsewhere over something like I-Light, the state’s high-speed, fiber-optic network for linking Indiana’s public and private college campuses, which Purdue and Indiana University jointly manage.

Hoffmann said dealing with video conferencing and collaboration is somewhat trickier because the raw material isn’t digital to start, unlike scientific data and visualizations, and must be converted. The researchers have done some testing of the system across the Purdue campus in West Lafayette. The next step is to try it between Purdue’s West Lafayette and Calumet campuses and over the Northwest Indiana Computational Grid, the scalable, high-speed and high-bandwidth system in Northwest Indiana designed for science and technology research and to enhance economic development in the region. ITaP and the Rosen Center, Purdue Calumet and Notre Dame are partners in the grid.

Calumet created a visualization lab this spring, funded through the Northwest Indiana Computational Grid, with a two-screen immersive virtual environment, haptic equipment to allow users to experience touch and feel when using the facility and supporting features, said Jack Moreland, visualization specialist at the Calumet campus.

Already, researchers at Calumet and steel industry partners have used it to step inside a blast furnace as part of a fluid dynamics simulation exploring ways to improve the manufacturing process, Moreland said. The resource also has been used in teaching chemistry, putting students in the middle of molecules they’re studying.

The system being developed by Hoffmann, Popescu and colleagues could open up numerous other possibilities for research and teaching at Calumet by making it easy to tap data collections and computing power at Purdue’s West Lafayette campus and elsewhere for visualization purposes, Hoffmann and Moreland said.

Writer: Greg Kline, (765) 494-8167, gkline@purdue.edu

Sources: Chris Hoffmann, (765) 494-6185, cmh@cs.purdue.edu Voicu Popescu, (765) 496-7347, popescu@cs.purdue.edu Jack Moreland, (219) 989-2765, morelanj@calumet.purdue.edu

Originally posted: December 8, 2008