More Power: Taking Education Up a Notch
By Amanda Cairo
Physics doctoral student Larz White needed more power. One computer just didn’t cut it for his research, so he looked for a solution. He found it in fellow physics graduate classmate Chris Mirabzadeh.
White’s research -- working with physics professor Francesca Sammarruca -- deals with microscopic predictions of nuclear matter properties and the process was taking too long to suit White.
“We have all these equations we code, and the computer can take seconds to hours to days when I work on a typical computer,” says White. “I wanted to cut that time down.”
White’s codes typically ran for a few days. White knew that his classmate Mirabzadeh was good with computers, and Mirabzadeh was intrigued by the challenge
“Larz knew I had a strong background and interest in computers,” says Mirabzadeh. “This project was a great opportunity to stretch my skills and build something completely new.”
What began as a “what if”, soon blossomed into a makeshift system and then became a supercomputer named Dirac, named after the noted physicist Paul Dirac, whom White admires.
In the beginning, Mirabzadeh and White bought junk computers from surplus to build the first prototype of Dirac. They set up a hodge-podged unit in the graduate student room with a constant fan hovering over it to make sure it didn’t overheat. The important fact, though, was it worked.
Then Mirabzadeh and White got serious. They purchased five high-powered computers and used them to build the supercomputer. Since the physics department helped them with purchasing, advice and support, the supercomputer is available for use in the department and is securely housed.
“The physics department has been very supportive,” says White. “Because of its support, we not only have this machine to use ourselves, we are able to offer it to others and help them gain new skills in parallel programming.”
Using a master computer and linking four nodes (computers) to it to divvy the workload, White’s computations can be run by parallel processing, to speed up the time it takes to run his computer code. Dirac’s speed has been pumped up to 120 GigaFlops (FLoating point Operations Per Second). A middle end computer averages 15-20 Gflops.
"It's a divide-and-conquer methodology," White says. "Every computer works together to compute a portion of the problem."
With the supercomputer, White’s research speeds up, and he’s able to do more, faster. It also helps further research at the University.
“The nuclear theory group at the University of Idaho develops models of nuclear systems starting from first principles. Computational facilities are an essential tool for our work,” Sammarruca. “The project put together by Larz and Chris will increase our computational power substantially and thus will be extremely beneficial to the research efforts of our group.”
And while Mirabzadeh doesn’t currently need the supercomputer for his research, the experience of building the supercomputer and using parallel programming was definitely time well spent. He also says that building a custom computer is becoming more common these days because it can save money, and it can be customized to the user’s needs. But Dirac was a little more challenging than your basic computer.
“Making several computers talk to each other and share a workload, that's the trick, that's the hard part, but that's what I'm good at and that's the part about building these types of systems that sets it apart and really takes advantage the capabilities of modern computers,” says Mirabzadeh. “This kind of project gives you a broader scope of what these computers can really do.”