RSS Feed (xml)

Powered By

Skin Design:
Free Blogger Skins

Powered by Blogger

Tuesday, July 01, 2008

Taking a byte out of research

ore supercomputers are showing up at Arkansas’ four-year universities, helping researchers tackle increasingly complex scientific problems.

The University of Arkansas at Fayetteville’s second campuswide supercomputer is among the world’s fastest according to the latest list, and UA’s Little Rock campus is preparing to install its first.

UA-Fayetteville’s “Star of Arkansas” is ranked 339 th on the International Supercomputer Conference’s 31 st Top 500 list.

That’s speedier than Harvard’s supercomputer, which made the list at No. 411. It’s also a better debut than UA’s first supercomputer, “Red Diamond,” ranked 379 th on the June 2005 list.

But with ever-changing technology and higher speeds, supercomputers don’t stay on the list long.

Star of Arkansas and Red Diamond use nearly the same amount of electricity, yet Star operates about eight times faster, at a rate of 10. 75 teraflops, said Amy Apon, a UA professor of computer science and computer engineering.

A teraflop means 1 trillion calculations per second.

“One of our researchers, Sergey Lisenkov, recently ran a job overnight on Star that would have taken a year [for a desktop computer ] to complete — if it could have even been done,” Apon said, referring to a physics research associate’s work that resulted in an initial model for future nanotechnology electrical devices.

“It took him 4 / 2 hours,” said Jeff Pummill, the university’s senior administrator for highperformance computing.

Fayetteville campus computer experts have been helping UALR prepare its new, unnamed supercomputer for installation.

“It is sitting here — we are just waiting for the space to be renovated,” said Srini Ramaswamy, chairman of the computer science department at UALR.

Pummill estimates UALR’s computer will run at 4. 35 teraflops, second-fastest in the state.

However, the ISC list’s No. 1 machine is leaving the terafloppers in the dust.

“Roadrunner,” housed at Los Alamos National Laboratory in New Mexico, became first to break the “petaflop barrier,” its owners announced two weeks ago. It runs at 1. 026 petaflops, or 1 quadrillion calculations, per second.

79, 000 WATTS AN HOUR About the size of eight extratall, extra-narrow, side-by-side refrigerators, Star of Arkansas has been operating about a month. Each of its eight units resembles the display racks of car-stereo components found in retail stores. Little lime-colored lights, blinking behind a black-matte grill with small round vents, don’t betray to visitors the mystery of what Star is up to at a given moment. Star consumes 79, 000 watts of electricity per hour, the equivalent of 790, 100-watt light bulbs. Small magnetic, dollar-store LCD thermometers affixed to Star show the air that its cooling fans expel in back to be 105 degrees. Targeted air-conditioning keeps the temperature at 60 degrees in front. “At about 115, I’ve told these guys to alert me,” Pummill said, pointing to two employees in a control booth who would summon Pummill to the scene like an emergency room doctor. He even has his own version of an ER “crash cart.” ‘COOL BOMB’ Sitting at his office desktop computer, Panneer Selvam logs on to user accounts for Star and Diamond. Selvam, who works in microelectronics, photonics and civil engineering, is searching his hard drive for a visual result he achieved for a pesky ongoing problem using a number of supercomputing experiments over several years.

He pulls up an animated model. The problem: How best to cool high-powered electronics, such as radar, weapons or laser systems.

He opens his palm to reveal a microchip. The goal is to build a little tank that will pump tiny water droplets into a coolingsystem chamber. The heat would convert the water into a cooling vapor for the chips before the vapor is sucked back out, condensed and recycled.

“I like to call it a ‘cool bomb, ’” Selvam said while pulling up a “movie” on his computer that resembles a cartoon with inanimate characters.

In the 3 D movie, he drops a 30-micron water droplet, colored red, on a surface of blue-colored water and water vapor.

“Chhhhh !” he says as he flicks a closed fist forward into open palm, mimicking a Japanese showman chef flinging water droplets on a hot grill. In the computer, the droplet creates a bubble under the surface that displaces the water and vapor, simulating the journey the cool bomb would make.

ASSIGNED JOBS Since Star is best at multipleprocess jobs, also called parallel tasks, Diamond, which should serve the university another two or three years, will be reassigned to do all the single-process, or serial, jobs, Apon said. She likened the parallel jobs to piecing together a quilt or giant jigsaw puzzle using many workers, each working their own sections that later would be combined into one. A serial job is more like one person looking at a single piece of data: “Like several friends each shopping for a car, but later comparing their data with each other,” Apon said.

So when researchers request their job, “They queue up,” she said.

Then, based on the kind of problem they assign and how many processors it will take, the jobs are assigned to Star or Diamond.

“There are lines for big jobs and lines for small jobs, like a reservation at a restaurant,” Apon said. “The party of 10 takes a bigger table.” Star was made possible by a partnership with Dell Corp. and an $ 803, 306 grant from the National Science Foundation.

The Fayetteville campus matched every $ 5 from the foundation with $ 1, said Collis Geren, the school’s vice provost for research and graduate school dean.

UA LR is ge tting about $ 700, 000 from two National Science Foundation grants and state matching grants for capital for its supercomputer and to build its “computing environment,” Ramaswamy said.

Albert Everett, computational specialist at UALR’s Graduate Institute of Technology, said installation of the supercomputer can’t begin until its server room is renovated, which will take an estimated 90 days. Setup and testing will take several months. Geren said supercomputers speed up research, save time and expensive lab materials, and increase the chances that scientists who use them will land future research grants. “You can model for a thousandth of the cost it would take to build a prototype,” he said.

SIMULATING PROBLEMS For certain complex scientific calculations, Pummill said, “We’ve left the generation where you can do the research in a lab to solve problems.” Some problems are too small, and others — such as the universe and its galaxies — too large for lab experiments. Other items under study, such as geological processes, which occur very slowly, and complex weather and climate questions, can be simulated much easier on a supercomputer.

“If you want to see what happens if the sun blows up, you simulate it,” Pummill said.

Doug Spearot, a UA-Fayetteville assistant professor of mechanical engineering, said supercomputers help him and his team at the nanoscale.

They study whether metals, plastics and polymers can be made stronger or more lightweight for the construction and automotive industries by learning how their building blocks behave at microlevel or nanolevel.

“We have powerful enough microscopes to see atoms,” he said.

But typically, researchers see only before-and-after snapshots of the atoms’ movements, because viewing movements in between would be difficult and expensive without supercomputer modeling.

“It can give us a real-time visualization of what’s actually happening during an experiment,” Spearot said.

No comments: