Scripps Institution of Oceanography
Seeing the Big Picture
The Technology Challenges
To cope with multiple and large data sets, visualization specialists face a variety of technology challenges, which fall into several broad categories:
- Category 1: Display issues related to building arbitrarily large “walls of information”
- Category 2: User interaction with data issues involving scaling, navigating, and simplifying image spaces intelligently and for overlaying, sequencing, slicing, and otherwise manipulating multiple data streams for the purposes of comparison
- Category 3: Networking issues related to moving data from one location to another at high bandwidths and controlling distributed visualization and collaboration processes
These categories are, of course, closely intertwined, so that research in one area often necessitates some development activities in the other areas as well. Such was the case with the iCluster project. Though nominally a “wall of information” project (Category 1), it was closely allied with an open source graphics middleware project (Category 2) funded by the NSF OptIPuter e-science networking initiative (Category 3).
“Linux running on consumer desktop machines was a possibility, but when you look at all of the costs involved, it’s certainly no less expensive than the Mac solution and if you factor in administration and porting costs, probably a good deal more expensive.”
Nayak’s former colleagues at the Electronic Visualization Laboratory, headed by Dr. Jason Leigh, developed the graphics middleware project — SAGE (Scalable Adaptive Graphics Environment). An open-source project, SAGE allows visualization systems developed on different platforms — and with display areas of widely different resolutions — to work collaboratively with the same data sets. Free SAGE downloads for Mac and other platforms are available from www.evl.uic.edu/cavern/sage.
“Our iCluster project tackled the problem of pushing more pixels out to tiled displays at very high resolutions,” says Nayak, who’s in charge of making it all work. “But even that relatively narrow slice of the whole visualization pie — ignoring the networking and graphics middleware issues — entails overcoming a number of specialized challenges, each of which is quite formidable.”
These challenges include the choice of displays, mounting hardware, computer hardware, and operating system. Other issues including how to drive the chosen display units, manage the data sets and workflows, and perform final image processing also need to be decided.
The Perfect Display
Having decided on a tiled array of flat-panel displays, the iCluster team needed to assemble the hardware and software components that would allow them to meet their design goals: 50 megapixels of total display resolution that can display flicker-free, full-motion video and computer-generated animation sequences.
With its 2560x1600 native resolution, brilliant color, sharp contrast, flicker-free operation, and its narrow bezel (which allows for arrangement in a grid with the picture areas interrupted by the narrowest possible “French window” effect), the 30-inch Apple Cinema Displays were the perfect solution. Moreover, the Mac platform met the project’s design requirements for performance and OpenGL graphics library support.
Graham Kent explains the geology of the Lake Tahoe region using the iCluster display.
The SIO Visualization Center had previously built systems using SGI and/or Linux backends. “SGI systems are terrific for very specialized applications but they’re not cost effective for something like iCluster,” says Nayak. “Linux running on consumer desktop machines was a possibility, but when you look at all of the costs involved, it’s certainly no less expensive than the Mac solution and if you factor in administration and porting costs, probably a good deal more expensive.”
Nayak adds: “Each entering graduate student here receives a 12-inch or 15-inch PowerBook. Most of our ocean-going research runs on Macs as well. Consequently, the computing infrastructure at SIO/IGPP is very Mac-centric and we wanted if possible to use a computing architecture for the iCluster that would fit in with the rest of our infrastructure.”
All of these factors argued in favor of an all-Mac system. To achieve the design goal of 50 megapixels in the main display area, the iCluster team settled on a tiled grid of a dozen 30-inch Apple Cinema Displays driven by a cluster of six Power Mac G5 headless display nodes, interconnected through a Netgear gigabit Ethernet switch. A seventh Power Mac G5, which serves as the master cluster node, has its own 30-inch Apple Cinema Display. The master node is the only one of the Power Macs equipped with a keyboard and mouse; the remaining systems are controlled from the master node using Apple Remote Desktop 2 and the open source Synergy software for sharing a mouse and keyboard across multiple desktops.