High performance computation has taken Mark Jones into the world of parallel reconfig-urable
computers - and he is making sure his undergraduate and graduate
students go with him.
"It's a one-of-a-kind super-computer," Jones said. "For certain applications, including image processing, data bases, and simulations, it has the potential to deliver high performance at a low cost." The machine is serving as a high-performance computing resource for several research projects and as a reconfigurable computing testbed.
"We're building the software to control the machine and allow users to take advantage of the processors, the configurability, and a high-speed network," he explained. The system was funded by the National Science Foundation (NSF) and Virginia Tech. Current ongoing projects are funded by DARPA.
An assistant professor in the computer engineering area, Jones brings a computer science background to his work, concentrating on programming and networking issues. In addition to the Tower of Power projects, he is involved in efforts to improve the programming of configurable computers, design and implement parallel algorithms for scientific computation, and develop Internet-based decision support systems for the Air Force.
One project in the Config-urable Computing Laboratory, with Lockheed Martin, involves developing and exploiting fast context switching field programmable gate arrays (FPGAs). "On a traditional FPGA, downloading one program into the device is very time consuming," he explained. "Traditional FPGAs are ill-suited to configurable computing where new programs are downloaded onto the FPGA in response to rapidly changing application requirements. On this particular project, the new FPGA from Lockheed Martin can do very fast context switching. We're figuring out how to exploit that feature to build software to help people construct applications that can make use of configurable computers."
"We're working to develop easier programming for configurable computers than what exists right now. We hope to increase productivity and open configurable computing to a wide range of programmers. Now there is a narrow range of programmers involved because of the difficulty level. Programming a configurable computer will still be more difficult than writing a C program, but we believe this effort is justified in those applications that require very high performance."
Jones and Athanas envision configurable computing eventually allowing users to dynamically upgrade/modify hardware to accommodate changes in their computing requirements, resulting in a reduction of computer hardware costs.
Jones actively involves students at all levels in state-of-the-art computing. In addition to actively recruiting undergraduates to work on the Tower of Power, he serves as faculty advisor to DISC, a virtual corporation that is developing medical information systems. He is also interested in developing the curriculum to encourage students to work with cutting-edge issues.
Last fall he taught a new course on configurable computing. This semester, he developed a new graduate-level course in parallel algorithms and applications. "We have the grad students using the Tower of Power cluster, an IBM SP2, and a network of workstations," he commented. "They are looking at algorithms that have applications in scientific computing and enterprise server applications."
His interest in parallel algorithms and applications is long standing. While earning his Ph.D. from Duke University, he worked on parallel eigenvalue problems and spent summers at NASA Langley in the Computational Structures Branch. At NASA, he was involved in analyzing the models for the space station and solid rocket boosters. "Analysis is computationally intense," he said. "We needed to use parallel computers, and I developed algorithms for using parallel and vector computers to solve vibration and buckling problems."
He then served for three years at Argonne National Laboratory, where he "worked on even bigger parallel computers. We were developing algorithms and software for using large parallel machines with over 500 processors to solve problems that arise in piezocrystal analysis and high-temperature superconductivity." Through his work at Argonne, he was awarded the IEEE Computer Society Gordon Bell Prize.
After Argonne, Jones spent four years as an assistant professor at the University of Tennessee, where his research involved looking at parallel computers, clusters of workstations, as well as configurable computers. While at Tennessee, he was honored with the NSF Career Award.
"The apparently insatiable appetite
for high-speed computing will continue to provide interesting
challenges for our lab in areas such as parallel computing and
configurable computing. Working with students to meet these challenges
in the classroom and the lab is a great working environment,"