LANS Informal Seminar: Michael McCourt and Hong Zhang

Seminar Title: 1) Progress in Parallel Implicit Methods for Tokamak Edge Plasma Modeling and 2) Sparse Triangular Solves for ILU Revisited: Data Layout Crucial to Be
Speaker: Michael McCourt and Hong Zhang,

Date/Time: 2010-10-27 15:00
Location: Bldg 240 Conference Center 1406-1407

Description:
There are two talks:

Talk by Michael McCourt
===============
Progress in Parallel Implicit Methods for Tokamak Edge Plasma Modeling

Performance of prototype tokamak fusion devices depends sensitively on
characteristics of the edge plasma between the hot core and surrounding walls. The edge plasma includes an especially wide range of physical time scales such that implicit numerical algorithms can substantially improve overall computational efficiency. This presentation introduces some of the benefits and challenges of parallel implicit solution strategies, with emphasis on preconditioned Newton-Krylov methods in the UEDGE and BOUT++ applications. We will discuss multiphysics issues in the context of the FACETS project, which is developing a multiphysics, parallel application to enable
modeling from the material wall to the plasma core. We also will explain how this fusion research is motivating new capabilities in the PETSc library to better handle strong coupling between two or more distinct PDE-based mathematical models.

Talk by Hong Zhang
============
Sparse Triangular Solves for ILU Revisited: Data Layout Crucial to Better Performance

A key to good processor utilization for sparse matrix computations is storing the data in the format that is most conducive to fast access by the memory system. In particular, for sparse matrix triangular solves the traditional compressed sparse matrix format is poor, and minor adjustments to the data structure can increase the processor utilization dramatically. Such adjustments involve storing the L and U factors separately and storing the U rows ‘backwards’ so that they are accessed in a simple streaming fashion during the triangular solves. Changes to the PETSc libraries to use this modified storage format resulted in over twice the floating-point rate for some matrices.