University of Rochester

Five New Supercomputers Push the Envelope at Rochester

October 24, 2002

In the center of a room humming with computer equipment at the University of Rochester stands one of five new IBM eServer supercomputers that will power a wide range of advanced research, from designing smart microprocessors to understanding how stars are born.

The largest of the five machines is operated by the Department of Computer Science. Using 32 microprocessors that are each faster than any in even the fastest desktop computers, researchers will work to develop power-efficient chips, methods to let programs run on machines of different kinds spread across the Internet, and more. The department's machine was acquired with the help of a $1.2 million gift from IBM's Shared University Research program, an equipment award program designed to promote research in areas of interest to both IBM and universities.

The other four supercomputers, each with 16 processors, are installed in the University's Laboratory for Laser Energetics (LLE), which houses the most powerful laser in the world. The machines will help researchers understand the process of fusion as the laser blasts millimeter-sized targets in a billionth of a second with more than 100 times the power used by the entire nation. By augmenting the laser shots with simulations on the supercomputers, researchers can tease out more information about the way fusion works, and how best to achieve it. Techniques for running these simulations have recently begun to take advantage of advanced multi-processor computer architecture like that in the IBM eServers, breaking down and distributing the tasks among the processors and providing the necessary brawn to evaluate projects that were thought too involved to attempt just a short time ago.

Researchers in the Department of Computer Science are using their new IBM eServer for a number of types of research, including collaborating with colleagues at IBM to simulate the operation of fast new low-power chips. A technique known as Complexity-Adaptive Processing reconfigures chips as they run, allowing them to meet the needs of software with as little energy as possible.

Working with colleagues in several allied fields, computer scientists are also exploring ways to improve the speed of the most demanding simulations and to allow programs running on machines of many kinds spread across the Internet to share memory as easily as they could if they were a single machine. In one example, software from the department's InterWeave project allows astronomers to animate the birth of stars on a desktop computer using the p690 located across the campus. Future work will use InterWeave as the foundation for virtual worlds in which humans and computers collaborate on complicated tasks.

Another project aims to make it easier for processors inside a supercomputer to synchronize their activities, which may lead to dramatic improvements in the speed of certain database systems and high-end operating systems.

The Department of Computer Science's machine was awarded to the University through a grant from IBM's Shared University Research (SUR) program, which provides computer technology to colleges, universities, and institutions of higher education around the world to facilitate research projects in areas of mutual interest. University researchers are also leveraging the IBM AIX5L UNIX operating system and other software products provided via IBM's Scholars program. The estimated value of IBM technology provided to the University exceeds $1 million.

The University has a long history of groundbreaking work in parallel computing, much of which has influenced today's multiple-processor computers. In the early 1980s, the Department of Computer Science was home to the largest shared-memory multiprocessor in the world.




Facebook