Monday, March 22
MS1
Parallel Multigrid Methods
10:00 AM-12:30 PM
Room: Ballroom A
Modern simulation codes must solve extremely large systems of
equations- tens, even hundreds of millions of equations. Hence, there
is an acute need for scalable parallel linear solvers, i.e.,
algorithms for which the time to solution (or number of iterations)
remains constant as both problem size and number of processors
increase. Multigrid, known to be an optimal serial algorithm, is
often scalable when implemented on a parallel computer. The speakers
in this minisymposium will discuss parallelizing multigrid solvers
for various problems and architectures. The machines range from the
high-end ASCI machines, with thousands of processors, to low-cost
clusters of workstations.
Organizer: Van Emden Henson
Lawrence Livermore National Laboratory
- 10:00-10:20 Parallel Semicoarsening Multigrid
- Jim E. Jones and Robert D. Falgout, Lawrence Livermore
National Laboratory
- 10:25-10:45 Transpose-Free Parallel ADI Methods in Multigrid
- Craig C. Douglas, University of Kentucky; Sachit
Malhotra, Morgan Stanley Dean Witter; and Martin H. Schultz, Yale University
- 10:50-11:10 Approaches to Parallel Multigrid with the Full
Domain Partition
- William F. Mitchell, National Institute of Standards and
Technology, Gaithersburg
- 11:15-11:35 Parallel Multigrid Solver of Unstructured Finite
Element Problems in Non-Linear Solid Mechanics
- Mark Adams, University of California, Berkeley
- 11:40-12:00 Design of a Multilevel Module for Parallel
Unstructured Grid Computations
- Karen Devine, John Shadid, Charles Tong and Ray Tuminaro,
Sandia National Laboratories, Livermore
- 12:05-12:25 A Parallel Implementation of Algebraic Multigrid
- Van Emden Henson, Organizer; Robert D. Falgout, Jim E.
Jones and Ulrike Meier Yang, Lawrence Livermore National Laboratory
LMH, 10/28/98, MMD, 11/16/98