The University Record, March 29, 1993

NSF, Ford fund center for high-performance computing

By Sally Pobojewski
News and Information Services

Researchers have received a $2 million grant from the National Science Foundation and a $1 million grant from Ford Motor Co. to create a new research center for high-performance computing.

The Center for Parallel Computing (CPC) will focus on the development and utilization of parallel computing technology for scientific and engineering applications. The University is supplying an additional $1.9 million in matching funds to establish the new center.

Parallel computing technology has the potential to outperform the most powerful supercomputers available today, and it is the focus of intense research efforts at university, federal and industry research laboratories throughout the United States.

“Today’s computers—everything from PCs to supercomputers—are based on serial technology where one processor completes the entire task from beginning to end,” says Randall L. Frank, director of information technology for the College of Engineering.

“The goal of parallel computing is to get hundreds or thousands of processors working together to solve a problem simultaneously by breaking it into smaller pieces,” says William R. Martin, professor of nuclear engineering and director of the center.

“It’s like building a house,” Frank explains. “You can finish the house much faster if you have many contractors working together, instead of just one contractor doing the entire job by himself.

“Effective parallel computing requires higher levels of coordination and communication, however,” Frank adds. “If all the contractors are standing around waiting for some concrete to be poured or getting in each other’s way, it could take longer to build the house than if one contractor did the job alone.”

Martin explains that parallel computing has the potential to fundamentally change the design process for automobiles, aircraft and virtually any product designed by engineers.

“By reducing the time it takes to solve a problem from days to minutes, parallel computing could allow engineers to design interactively, explore more design alternatives, use more complex simulations, and work within the shorter product development cycles industry faces today,” he says.

Parallel processing also has potential applications in fields as diverse as pharmaceutical drug design, mapping the human genome, radiation cancer therapy and space plasma physics.

“Ford Motor Company has long recognized the importance of large-scale computation in reducing the time required to produce quality products,” says Wayne Hamann, Ford Motor Co. director of product and manufacturing systems. “Collaboration with the U-M Center for Parallel Computing will help maintain a leadership role for Ford in applying leading-edge supercomputer technology to automotive problems.”

Fourteen researchers from various disciplines will merge their expertise in the center. Their goal is to develop and implement the new algorithms, software and communications protocols required to make parallel computing a useful tool for computational scientists.

“Programming many processors to work together is much more difficult than programming the most advanced vector supercomputer today,” Martin explains. “It’s basically a communications and synchronization problem. Our first task will be to determine which problems lend themselves to parallel processing and how best to partition these problems into their component tasks.”

“The center will establish working partnerships between computer researchers and users from many disciplines within the University and from industry,” Martin says. “We believe these partnerships hold the key to the future of computing.” The center will provide a unique opportunity to use this key to help unlock the full promise of this exciting new technology.”

The first phase of the project is already under way, using a newly acquired 32-processor Kendall Square Research computer. During the summer of 1993, the

U-M plans to add a second parallel machine to the center, according to Frank.

“We also have in place a loosely-coupled parallel facility consisting of IBM RS/6000 workstations and plans to enhance the network connecting these systems to allow them to communicate significantly faster,” Frank notes.

Working with several different hardware systems will allow researchers to determine which type of parallel computer architecture is most suitable for specific types of problems, Frank explains.

“The three hardware systems to be available at the center represent the three distinct ways of solving communication and coordination problems in parallel systems,” Frank added. “The Center will give researchers an opportunity to experiment with the full range of potential solutions to the problem.”

The U-M’s $1.9 million contribution to the center consists of equipment funds, technical expertise and staff support. It is being provided by the College of Engineering, the Information Technology Division (ITD) and the Office of the Vice President for Research.

In addition to Martin and Frank, principal researchers on the project include: Rayhaneh Akhavan, assistant professor of mechanical engineering and applied mechanics; Daniel E. Atkins, dean, School of Information and Library Studies, and professor of electrical engineering and computer science; Edward S. Davidson, professor of electrical engineering and computer science; Tamas I. Gombosi, professor of atmospheric, oceanic and space sciences and of aerospace engineering; Trevor N. Mudge, professor of electrical engineering and computer science; Kenneth G. Powell, assistant professor of aerospace engineering; and Quentin F. Stout, professor of electrical engineering and computer science.