NPAC Technical Report SCCS-656

An Alternative to Data Mapping for Parallel PDE Solvers: Parallel Grid Generation

Nikos Chrisochoides

Submitted October 1 1993


Abstract

In this paper we identified and outlined the disadvantages of the traditional data mapping methods for the numerical solution of PDEs on distributed memory MIMD machines and we proposed a new approach that eliminates some of the disadvantages. Specifically, we presented a data-mapping approach based on parallel structured grid generation. The new approach is based on composite block structures to contract the size of the data-mapping problem. It is ten times faster than the fastest traditional data-mapping method, for relatively small problems, and approximately O(P) times faster, for very large problems (i.e., millions of grid points) that are processed on coarse-grain distributed memory MIMD machines with P processors.


PostScript version of the paper