News from PRINCETON UNIVERSITY
Ordinary PCs create high-performance computing
PRINCETON, N.J. -- The Office of Information Technology has announced plans to build a high-performance computer system that will be available this spring to members of all departments for teaching and research.
The computer system, a type known as a Beowulf cluster, will consist of 32 off-the-shelf personal computers connected with a high-speed network. This setup offers many benefits of custom-designed scientific supercomputers at a fraction of the cost, said Curt Hillegas, OIT's manager of research and academic applications support. OIT is building the system in partnership with the Dell Computer Corp., which is providing and installing the computers at a reduced cost.
The computer system could help scientists simulate anything from the early history of the universe to complex chemical reactions to demographic trends.
The initiative grew out of recommendations of the Research Computing Advisory Group, a committee of faculty members, OIT personnel and departmental computer support staff, said Hillegas. "We are trying to respond with services that the faculty said would really support their research."
The new computer system is expected to serve three main purposes:
In the last few years, at least 11 research groups on campus, in such departments as geosciences, astrophysics and mechanical and aerospace engineering, have built their own computing clusters of computers, most of which use commonly available PCs.
"This is a way of spreading the technology to a broader range of departments," said Bjorn Engquist, the Michael Henry Strater University Professor of Mathematics. Engquist directs the Princeton Institute for Computational Science and Engineering, which researches and supports science involving high-performance computing and which worked closely with OIT in designing the Beowulf system.
Increasingly, said Engquist, scientists in all disciplines rely on simulating real-world phenomena on computers and testing their understanding of the subject by manipulating the computer program. "Computational science is an increasingly important paradigm, often connecting theory and experiment," he said. Understanding the properties of a large system, such as the Earth's atmosphere, often requires calculating the interactions of billions of components, which creates an enormous demand for computing power.
In the past, scientists needing powerful scientific computing turned to expensive supercomputers. Beowulf clusters, which were invented by NASA engineers in 1994, take advantage of the dramatic improvements in computing price per dollar in the PC industry, said Engquist. The Princeton Institute for Computational Science and Engineering, which began functioning last fall, provides its own general-use computing cluster, which is smaller than the Beowulf cluster.
The Beowulf initiative also complements the work of the Program in Integrated Computer and Application Sciences (PICASSO), a graduate training and research program focused on integrating techniques of computer science with other disciplines that are becoming more reliant on computation. Beowulf clusters are emerging as a common form of high-performance computing, said J.P. Singh, associate professor of computer science and director of PICASSO. "So this tool will allow students to get hands-on experience writing, debugging and optimizing parallel code for a very real and important platform,” said Singh.
"This initiative represents one step in our ongoing effort to listen to the needs of faculty members, students and administrators and respond with services that benefit as many areas of teaching and research as possible," said Betty Leydon, vice president for information technology. "We are grateful for the strong support from faculty members in developing the Beowulf cluster and look forward to working with many more people in putting it to use."
The OIT Beowulf cluster will be powered by 32 Dell PCs, each containing
two 2.4-gigahertz Intel processors. The computers will run the RedHat
Linux operating system. The University recently received a shipment of
the PCs and expects the cluster, which will be located in the OIT building
at 87 Prospect St., to be built and running this spring. More information
is available at http://www.Princeton.EDU/~raas/beowulf.
Researchers who want to use the system should contact Hillegas at curt@Princeton.EDU