![[Case Western Reserve University -- Toolbar]](/pix/lowpro.gif)
Posted 12-10-99
CLEVELAND -- Not long ago, the idea of building your own supercomputer would have seemed outlandish. But several CWRU students did just that by building the Beowulf Supercomputer Cluster at Case Western Reserve University. The cluster provides access to high-power computing without steep costs.
Ordinarily, researchers who want to do some serious number crunching must link up with supercomputer centers like the ones in Pittsburgh and Columbus, according to Brandon Palmer, manager of the Smith Computer Lab at CWRU. These centers have extremely powerful mainframe computers that can manipulate vast amounts of data in a brief time. Unfortunately, it can be costly to buy time on these computers.
Instead the CWRU team of "do-it-yourselfers" has combined a cluster of 15 inexpensive personal computers and networked them together so they work like a supercomputer. The project refers to a group of supercomputers built from off-the-shelf hardware using ATM (asynchronous transfer mode) for the network interface.
"Many people are interested in getting access to powerful machines that can do the number crunching they need, but few can afford that kind of power," Palmer said. "The Beowulf Project solves that by using clustering technology to turn a large number of smaller, off-the-shelf machines into a large and robust system. We can put a great deal of computing power into the hands of people who would not have any other way to get access to it."
The Intel-powered personal computers are connected by software and hardware. Ray Neff, vice president of information services at CWRU, supplied the ATM switch for the cluster, which allows the computers to "talk" to each other.
The CWRU cluster is patterned after the Beowulf Project started by NASA's Goddard Space Flight Center in 1994. The first Beowulf machine focused on problems related to the large data sets involved in Earth and space science applications.
NASA researchers dubbed their machine Beowulf after the hero of the eighth century epic poem who slays a monster and defeats a dragon who is ravaging his kingdom. Like the mythic hero, the Beowulf Project founders opted to master the technology and build their own machine.
The Beowulf Supercomputing Cluster at CWRU is an independent study project by a group of CWRU undergraduate and graduate students, with Palmer spearheading the effort.
The team members from the Department of Electrical Engineering and Computer Science are sophomore Chris Connelly, senior Jay Hart, and juniors Dave Monk and Paul Saitta. Jose Nazario, also a team member, is a graduate student in biochemistry. Frank Merat, associate professor and interim chair of electrical engineering and computer science, is the group's advisor.
When the Smith Lab received a donation of new computers from the Case Alumni Association, Palmer grabbed some of the Dell Intel Pentium 120 computers that were being replaced for the cluster. The cluster has 14 nodes, data entry points in a database management system. The cluster has the computing power of about 7 gigahertz and uses a generic software language called PVM, which stands for parallel virtual machine.
Nazario is already using the cluster for his research in molecular dynamics simulation.
"This allows me to look at very large biomolecules such as proteins and see how they move. It helps me to understand how the enzymes function like proteins," Nazario said.
CWRU faculty and students will be able to use the Beowulf system this fall if they have a demonstrated need, but users will be required to schedule time on the cluster, according to Palmer. To date two faculty have signed on to use the cluster for projects involving particle physics and to simulate protein structure.
The cluster will be used for parallel programming. This means that several problems can be performed simultaneously by different computers, or concurrently by the same unit, saving time and energy.
"The simulations I run can only be run in parallel," Nazario said.
With parallel processing, the server organizes the task then hands it to the nodes to do the work. The nodes return completed portions of the task to the server, which then reassembles the information and does the work that needs to be done, he explains.
Palmer said that the students who worked with him to develop the cluster gained more than just course credit. The project was a real educational opportunity, he noted, as he recalled a comment by one of his team members.
"At one point Dave Monk turned to me and said, 'This is why I came to CWRU -- because you get to do things like this. You get experiences that are hard to find other places.' "
For more information about the Beowulf Supercomputer Cluster at CWRU, visit http://home.cwru.edu/beowulf/.
![[Toolbar]](/pix/ribbon.gif){kind=small}