NPAC Technical Report SCCS-674

Parallel Structured Grid Generation

Nikos Chrisochoides, Animesh Chatterjee, Vaidyanathan Rajani, Geoffrey Fox

Submitted March 30 1994


Abstract

Partial Differential Equation (PDE) solvers. In this report we present our progress on modules for parallel algebraic and elliptic grids, a parallel multi-block Euler discretization module, and parallel iterative solvers. A lso we define the interfaces between : (1) parallel grid generation modules and multi-block Euler module, and (2) multi-block Euler module with parallel iterative solvers. Our approach for the solution of the data-mapping problem reduces the employment of sequential data pre-processing required for the data-parallel PDE solvers and at the same time exploits the re-usability of existing well written and tested sequential structured multi-block methods for parallel CFD codes. Preliminary data indicate that this approach is ten times faster than the fastest traditional data-mapping method, for relatively small problems (i.e., tens of thousands of grid points) and approximately P times faster for very large problems (i.e., millions of grid points) that are processed on coarse-grain distributed address space MIMD machines with P processors. The development phase of the parallel Algebraic and Elliptic grid generation modules, multi-block Euler discretization module and interface with Parallel Iterative Methods should be completed by the end of May. The interface of grid modules with NGP's\footnote{National Grid Project at MSU} and EagleView's geometry modeler, and data mapper as well as the extensive evaluation to various parallel platforms, and documentation will be completed by the end of this year.


PostScript version of the paper