Imagine trying to find Earth in a photograph of the entire universe on a cell phone screen and you'll have an inkling of the task facing an IT team helping biomedical researchers view huge datasets on a PC.
"Trying to view these datasets on a typical monitor can be analogous to looking through a straw into the haystack," said David Lee, BioWall Application Engineer for National Center for Microscopy and Imaging Research (NCMIR) and the San Diego Supercomputer Center (SDSC) at the University of California, San Diego, Calif.
One half of NCMIR technology development facility houses scientists, the other half the IT group, which develops life sciences applications, imaging technologies and systems for remote instrumentation control and exploring visualization methods. Overall, NCMIR's 70 employees' expertise covers biology, mathematics, physics, computer sciences to mechanical engineering.
Over 30 servers are used by NCMIR, and they run Linux, Windows, Solaris and Mac OS on various architecture types and two clusters, one 64-bit and one 32-bit. "Most scientific applications are now being developed on Linux due to its flexibility, available software, ease of deployment, cluster operating systems such as [open source project] Rocks and management tools," said Lee. The Rocks Linux distribution is a high-performance cluster toolkit.
At NCMIR, Lee's IT team has modified instrumentation and written applications to take large images from light and electron microscopes. Unfortunately, the size of these images is far greater the resolution of typical computer displays.
"Even with the high resolution tiled display, we are only able to view a fraction of the entire dataset simultaneously," Lee said. Researchers' productivity is reduced because, on a small screen, they can't detect the slight grey-scale changes that distinguish the useful data in electron microscope images.
For the past few years, NCMIR scientists have been frustrated with computer technology, particularly digital imaging. Many have gone back to using older recording mediums, such as film.
Frustrated himself, Lee began to search for a way to 'supersize' a computer display.
First, Lee's team evaluated IBM T221 displays. They worked well for viewing very large amounts of data. "Because of the physical size of the displays, they are only useful for a single person sitting directly in front of the displays," said Lee. Most of the work on the biological side is collaboration, so a one-person display didn't fit the bill.
Then, the IT team took a cue from the type of postmodern art installations that display images on walls with banks of video screens. They created a tiled wall, the BioWall Tiled Display, of high-resolution, flat-panel displays placed five across in four connecting rows. The 40-megapixels display shows detailed two- and three-dimensional images, creating more realistic and easy-to-see depictions of biological structures and cells and tissues in relation to each other.
Lee's team ran into no problems installing the BioWall's 21 Dual Opteron Sun Java Workstations with 8GB of RAM, NVIDIA Quadro 3000G graphics cards, dual- gigabit Ethernet and a single 10k RPM SCSI drive.
Rocks Linux plays an important role for BioWall. "The operating system was deployed with the Rocks mechanisms with a special package to make Rocks clusters suitable for graphics and multi-display clusters," said Lee. "Administration requirements are very low with Rocks clusters, requiring only the occasional maintenance from systems administrators."
The BioWall was a hit with the scientists, even the recalcitrant biologists who had clung to using older film technologies. Productivity increases have been dramatic. The only shortcoming of BioWall, Lee said, is that "it is made of many small pieces and is not easily portable." That's a hint to display manufacturers, and visions of a single-paneled BioWall probably dance through Lee's dreams.
Rather than dwell on dreams, however, Lee has concrete plans for adding a 3D overlay on top of the BioWall displays. Also, he's evaluating motion tracking devices that would increase biologists' interactivity, enabling them to experiment with usability scenarios with other input methods for the display.