Aztec: A parallel iterative package for unstructured Grid Problems
R. S. Tuminaro, S. A. Hutchinson, J. N. Shadid
Sandia National Laboratories
Albuquerque, New Mexico 87185
Abstract
We describe a parallel linear system package. This package incorporates a
number of novel iterative methods and preconditioners for the solution of
linear systems. In this talk we emphasize the parallel programming ease
associated with this package even for people who have little parallel
computing experience. In particular, a software tool is described that
greatly simplifies the parallelization process. The tool transforms a
serial matrix description into a distributed matrix suitable
for linear algebra kernels used in the iterative solver package.
Once the matrix is transformed, AZTEC can be used on the nCUBE 2, the
IBM SP2, and on the Intel Paragon.
The talk will describe the software tool as well as summarize the
iterative algorithms and preconditioners that are available. The major
focus of the talk will be to illustrate the significant simplification
in parallel programming effort that results from using this software.
Additionally, we show that these programming simplifications do not
result in a loss of efficiency. In particular, the time for the
transformation tool is insignificant and that the iterative methods
make good utilization of the machine (i.e. high Gigaflop rates are
possible even for sparse matrix applications). Numerous code examples
(structured 2D & 3D, unstructured finite element, high order stencils)
will be given to support these points.