Applied Computational Mathematics Division


Up

Parallel Adaptive Scientific Computation in Heterogeneous, Hierarchical, and Non-Dedicated Computing Environments

James D. Teresco
Williams College, Department of Computer Science

Thursday, June 15, 2006 15:00-16:00,
NIST North (820), Room 145
Gaithersburg
Thursday, June 15, 2006 13:00-14:00,
Room 4511
Boulder

Abstract: Large-scale computation has become an essential tool in many fields of science and engineering. Such computations are now being performed in parallel computing environments ranging from small clusters to large supercomputers. These systems can be comprised of nodes with different numbers of processors and include different processor speeds and memory and communication capabilities. All parts of the system may be subject to shared usage. This talk will first discuss the types of computations we consider---adaptive computations that require a dynamic rebalancing of data. Traditional dynamic load balancing procedures and tools will be discussed. The talk will discuss possible approaches in heterogeneous, hierarchical, and non-dedicated computing environments, focusing on two tools that we have developed: the Dynamic Resource Utilization Model and the Zoltan Toolkit's hierarchical partitioning and dynamic load balancing procedures.

Speaker Bio: Jim Teresco is an Assistant Professor in the Department of Computer Science at Williams College. He received his BS in Mathematics and Computer Science in 1992 and his MS in Computer Science in 1993, both from Union College. He received his PhD in Computer Science from Rensselaer Polytechnic Institute in 2000. He also spent several months at Sandia National Laboratories as a Visiting Faculty Researcher in 2003 and 2004. He has been awarded a Founders Award and the Robert McNaughton Prize by Rensselaer, and the Martin Terry Resch Prize and the Computer Science Alumni Silver Award by Union College. He is a member of ACM, the IEEE Computer Society, USACM, SIAM, Sigma Xi, and Upsilon Pi Epsilon. His research interests are in parallel, adaptive scientific computation in heterogeneous, hierarchical, non-dedicated and transient computing environments.

Presentation Slides: PDF

Contact: W. F. Mitchell

Note: Visitors from outside NIST must contact Robin Bickel; (301) 975-3668; at least 24 hours in advance.