NMSU branding

New Mexico State University

New Mexico State University

News Center




NMSU profs build parallel computer, invite researchers to sample

How can you get supercomputing power in your own backyard for one-tenth the cost of traditional machines? Build it yourself, answered New Mexico State University computer science and electrical engineering professors. With help from a $75,000 National Science Foundation research instrumentation grant, they built a high-performance supercomputer from scratch.


Researchers began clustering workstations together to work on single problems in the mid-1980s; however, the NMSU supercomputer belongs to the Beowulf class, a concept which was developed in 1993, said Srinivas Aluru, assistant professor of computer science and director of the Parallel Computing Laboratory. He built the computer, named Medusa, from 32 interconnected personal computers of the type you might find in any office or home.

This process resulted in a tremendous cost savings. A traditional parallel computer might cost about one million dollars, Aluru said, but the NMSU team spent less than $130,000 to build Medusa. And they can use software that's made available for free, he said. Now they look forward to seeing other NMSU researchers take advantage of the computing power available on their own campus.

The 32 PCs used for Medusa are very fast computers with extra memory connected by a high-speed network. The network is the most important part of the system, Aluru said. It is interconnected so that many pairs of computers can communicate simultaneously. It transfers messages back and forth between the computers as quickly as 1.28 Gigabits per second. "It does no good to have fast computers without a fast network," he said. "That is what makes this system a parallel computer."

Parallel computing uses many computers together to solve problems that would take too long or are too complex for a single machine. This gives researchers an advantage, Aluru said. "The traditional ways to approach research are with theory and experimentation," he said. "In the last decade we have added computation."

He explained that in some instances researchers cannot perform experiments because they are too expensive, prolonged or impractical. For example, astronomers may be researching how the universe develops or physicists might be exploring how atomic bombs explode. In these cases they can use complex computer programs to determine results, Aluru said.

He is using Medusa to work on factors determining the three-dimensional structure of protein molecules. The research is important, he said, because the structure of protein molecules determines their function in living organisms, and experimental procedures to discover the structures are complex, expensive and extremely slow.

Using parallel processing for this purpose is called computational biology, Aluru said. He believes it is just one example of the interdisciplinary opportunities brought about by parallel computing. His fellow researchers agree.

Steven Castillo, a professor of electrical and computer engineering at NMSU, uses Medusa to solve practical problems in electromagnetics. That includes modeling and designing antennas to minimize electromagnetic interference, a phenomena many people witness when they turn on their vacuum cleaner and get lines on their television, he said.

Eric Johnson is also an NMSU electrical and computer engineering professor, but when he uses Medusa to solve problems he throws away the answer and records how hard the machine worked to get it. He is exploring how diverse supercomputing designs perform with various workloads. Johnson is evaluating the Beowulf system "to see how hard we can push it compared to more traditional high-performance computing," he said. His goal is to determine which categories of parallel computing applications will work best with this type of parallel computer.

All three researchers believe it is a great advantage to have this computing power on NMSU's campus. "Even though they say you can have access to supercomputing elsewhere, you still have to ship your problems there," said Castillo. "Sometimes that takes longer than solving your problem. Then they have to ship the answer back to you." Having Medusa on campus avoids those problems as well as eliminating the wait while others use the resources, he said.

Aluru said parallel computing can benefit many fields, including various engineering disciplines, physics, astronomy and others that use computational research. The interdisciplinary approach can be a slow process, Aluru said, but he is optimistic his colleagues will realize the resource that is now available and use Medusa for their research. "We are bringing parallel computing within reach of many faculty," said Aluru. "A lot more can easily do research now."