« Distinguished Lecture Series on Internet and Grid Computing       High-Performance Computing in the 21st Century       Monday, March 31, 2003   ¤   11:00am-12:00pm Coffee: 10:30am ACES 6.304 (Seminar room) Daniel Reed     reed@ncsa.uiuc.edu Professor, Department of Computer Science Director, National Center for Supercomputing Applications (NCSA) & National Computational Science Alliance (Alliance) University of Illinois, Urbana-Champaign   Abstract   A new infrastructure for computational science is emerging that promises to transform the way science and engineering research are conducted. Based on core computer and computational science research, commodity processors, open source software, inexpensive storage devices and high bandwidth networks, computational Grids will soon distributed scientific instruments, computing systems, and collaborative groups. This talk focuses on two issues: the computer science challenges and software infrastructure needed to support parallel and distributed Grid applications and systems the computational infrastructure and applications behind the NSF TeraGrid. The latter is the anchor for NSF's cyberinfrastructure, connecting NCSA, SDSC, PSC, Caltech and Argonne in support of next-generation scientific applications. I will also sketch some visions of a possible "Renaissance" collaboration among computing technologists, scientists and engineers and the arts and humanities.   Biography   Professor Reed's primary research interest is in exploring performance optimization techniques for large-scale parallel, distributed, and Grid-based computing systems. A large portion of his research has focused resource management policies that dynamically adapt to changes in resource availability and application behavior. As part of this work, his research group has characterized parallel input/output access patterns explored file system policies suitable for systems with hundreds or thousands of disks. His research group developed the Pablo© suite of scalable performance tools, a software environment for instrumenting terascale applications, analyzing the resultant performance data using statistical and graphical tools, and gauging appropriate optimization measures. Pablo is used to tune parallel applications, study input/output bottlenecks, and analyze I/O access patterns and server performance. Several high-performance computing vendors have commercialized Pablo tools. Recent research extensions to Pablo support performance contracts to contracts guide run-time environments in configuring object programs to available resources and in deciding when to interrupt execution and reconfigure to achieve better performance. As part of this research, the Pablo group is defining and integrating fuzzy rule sets to express expected levels of performance and investigating adaptive, closed loop file system policies that dynamically change file system caching and prefetching policies to maximize performance. Statistical, Markov and time series models are being used to describe system behavior as input to file system and resource management policies. The goal of the graphical, immersive visualization environment is to make software developers active participants in the virtual worlds of their executing software, rather than mere spectators. In the physical world, humans sense and react to their environment, changing it either directly, via their actions, or indirectly, using tools that expand the range of potential intellectual or physical actions. Successful virtual environments for performance data display and software manipulation provide equally powerful and intuitive interfaces. While he maintains this active research effort in the Department of Computer Science, Reed's primary professional role is as the Director of the National Computational Science Alliance (Alliance) and the National Center for Supercomputing Applications (NCSA) at the University of Illinois, Urbana-Champaign. In this dual directorship role, Reed provides strategic direction and leadership to the Alliance and NCSA and is the principal investigator for the Alliance cooperative agreement with the National Science Foundation.   He is one of two principal investigators and the Chief Architect for the NSF TeraGrid project to create a U.S. national infrastructure for Grid computing. Professor Reed received his Ph.D. from Purdue University in 1983. In 2000, he was named Edward William and Jane Marr Gutgsell Professor. Additional honors and awards include IBM Faculty Development Award (1984); NSF Presidential Young Investigator Award (1987) Beckman Associate, Center for Advanced Study Associate (1989); University Scholars (1989); and Xerox Award for Faculty Research (1991). « 2/5-6/03 Raghavan