Research Supercomputing: Linking Chains of Knowledge through the Grid
Supercomputing in the New Millennium
Let's think big . . . of distances so vast across the universe that we measure them not in miles or meters, but in light years. Out there among the quasars and mysterious black holes, billions of light years from Earth, we continue to search for the answers to ancient questions that have kept the human fires of curiosity burning. How was the universe created? How do we fit into this immense cosmos?
Let's think small . . . of a world so small that its size is measured in nanometers, billionths of a meter, units so infinitesimal that a million nanometers could crowd into the width of this letter "I." We continue to seek to understand the properties and rules of this tiny world, too, searching for answers to questions that intrigue and perplex usquestions of how we can manipulate atoms to improve the materials and processes of our everyday lives.
These frontiers are vastly disparate. But exploring them, unlocking the mysteries and the potential they hold, requires precisely the same toolsmassive computational power and a rapid-fire network linking scientists in the common cause of discovery.
World-Class Research Begins with World-Class Information Systems
The University of Kentucky signaled its intent to be a national leader in high-performance computing in 1987 with the purchase of its first supercomputer and the establishment of the Center for Computational Sciences. The center currently serves to encourage new and innovative uses of computers; provide an optimum environment for training and for development of programs; support interdisciplinary projects where the computational expertise of one discipline can be transferred to another; initiate collaborations with outside research through the support of visitors, seminars, workshops and conferences; and test state-of-the-art hardware and software.
The second supercomputer to take up residence at the center was the Hewlett-Packard (HP) Exemplar, a high-performance scientific computer. Within 18 months of installation, the Exemplar increased UK's high-performance computing capabilities to 99 Gigaflops (one Gigaflop equals a billion floating-point computations per second).
The latest demonstration of UK's commitment to supercomputing excellence came with the acquisition of two supercomputers from Hewlett-Packard. The HP Linux cluster has 248 3.4GHz Intel Xeon em64t processors with 2GB RAM per processor (total of 512GB RAM across all nodes), a total of 8 Terabytes of high-speed disk storage, and has Myrinet high-speed message-passing interconnect for internode communication.
The second purchase, the HP Superdome, has a total of 256 Itanium-2 processors and is connected with the high-speed message-passing interconnect for internode communication.
"In carrying out UK's research mission, it's essential for faculty to have the very best supercomputing tools available. Some of the difficult questions our researchers ask can be answered only through supercomputing processes."
Wendy Baldwin, UK Executive Vice President for Research
Both the HP Linux cluster and the HP Superdome have a peak of 1.7 Teraflops capacity (one Teraflop equals a trillion computations per second).
"Both platforms are ranked in the top 500 supercomputers in the world, and the total capacity places the University of Kentucky in the top 10 academic centers in the country," says John Connolly, who has served as director of the Center for Computational Sciences since its inception.
"Hewlett-Packard is committed to providing affordable and reliable cycles at the top end of the computational pyramid," says Winston Prather, vice president, HP High Performance Technical Computing. "The University of Kentucky, which for many years has been the premier academic HP site in the United States, is an ideal place to test the HP supercomputer architectures. We are really pleased that the university has chosen to install these two top-end HP machines. The research done utilizing them and how these machines perform will tell us a great deal about how the HP vision fits with the computational research community. We look forward to an innovative and productive partnership with UK."
The University of Kentucky utilizes its network for all facets of the university community, according to Doyle Friskney, associate vice president for information technology and chief technology officer. "Students use the network to find information worldwide. Faculty utilize the network connections for scholarly research, to maintain contact with peer researchers, and to explore new research opportunities." Faculty have used the network connections to access everything from library resources to laboratory equipment and specialized research facilities. Administrators use the network connections for contact with vendors and partners in managing the university and the hospital. The network has become the common thread at the university in managing cross-campus communications.
Going for Distributed Computing with Like-Minded Partners
"Computational grids integrate high-performance computers, advanced visualization environments, remote instruments, and massive databases via high-speed computer networks."
John Connolly, Director, UK Center for Computational Sciences
UK's leadership role in supercomputing was solidified in the past seven years through a partnership between the university and the National Computational Science Alliance, established through a grant from the National Science Foundation. This coalition of computational scientists and educational professionals at more than 80 universities across the country established, by the end of the millen-nium, crucial computational and information infrastructure.
The Alliance, says Connolly, also gave rise to what he considers the future of supercomputingdistributed computing. "Distributed computing aims to promote the development and advancement of technologies that provide seamless and scalable access to wide-area distributed resources," says Connolly. "Computational grids integrate high-performance computers, advanced visualization environments, remote instruments, and massive databases via high-speed computer networks."
The distributed grid is a network. The idea is analogous to an electric power grid, where power generators are distributed, but the users are able to access electric power without worrying about the source of energy and its location.
At UK, GridChem is a new consortium involving the University of Kentucky as the lead university partnering with the University of Illinois at Urbana-Champaign, Louisiana State University, the University of Texas, and Ohio State University. The National Science Foundation’s National Middleware Initiative is funding the consortium in the amount of $3 million over three years.
"During the Alliance, we became the hub for doing particular kinds of calculations that focused on computational chemistry. The GridChem is following up on that by building a distributed network just for chemistry," Connolly explains. "Specialized computational centers, each focused on a particular type of calculation, are the wave of the future."
Middleware to Facilitate Easy Access
One problem with distributed computing is that each machine in the chain has its own system, its own "look." This is where middleware comes in, the software between the user and the grid. Middleware is a sort of common denominator that makes the grid user-friendly.
Connolly explains: "Suppose a researcher here is working on a calculation, but the Kentucky machine is full or busy. She may then try another machine, the Illinois computer, let's say. But it has a different operating system, and a different log-in system. Without middleware, the researcher would have to remember 20 different passwords for the 20 different machines on the grid. Middleware makes them all look the same." UK is taking the lead in developing the middleware needed to give the grid a uniform look.
The Computational Center and National Security
In addition to serving as a tool for scientific research, supercomputers at the University of Kentucky are now playing a role in national security. The U.S. Treasury Department is concerned about the nation's computer networks, especially the infrastructure for the banking industry. The department is charged with protecting the banking industry from terrorist attacks, which could take the form of attacks on the cyber-infrastructure.
"In the new GridChem middleware project, we're working with some of the same researchers that we collaborated with for seven years during the National Computational Science Alliance project. We know their capabilities. It helps that we all know each other's talents. We can be productive right out of the starting gate."
Barbara Kucera, Deputy Director, UK Center for Computational Sciences
"Either physical or cyber attacks could potentially do significant harm to the country's banking industry, could disrupt America’s financial infrastructure," says Jim Griffioen, an associate professor of computer science at UK. "The project we’re involved in, which is funded by the Treasury Department, is focused on thoroughly understanding the existing system in order to be able to pinpoint any weaknesses and vulnerabilities. Then we’ll propose solutions to fix those vulnerabilities."
Griffioen says that UK was tapped for this project, with the University of Louisville and two technology companies as partners, because UK has an established laboratory for advanced networking and a strong group of networking researchers. "We have experts here who know how to build robust networks," Griffioen says,"researchers who know how to build networks that can continue to operate despite natural disasters or attack, and provide services. We have expertise to offer those in charge of cyber security."
The UK team working on this project includes Connolly, Griffioen, Barbara Kucera, deputy director of the UK Center for Computational Sciences, Rafael Finkel, professor of computer science, Mukesh Singhal, professor of computer science and Gartner Group Endowed Chair in Networking, Zongming Fei, assistant professor of computer science, and Ken Calvert, associate professor of computer science.