DISTRIBUTED-NETWORK PARALLEL COMPUTING FOR ADAPTIVE METHODOLOGIES IN COMPUTATIONAL FLUID DYNAMICS

A GENERAL-PURPOSE, SOLUTION-ADAPTIVE, FINITE-ELEMENTS CODE THAT TREATS BROAD CLASSES OF INCOMPRESSIBLE FLOWS HAS BEEN DEVELOPED. THE CODE EMPLOYS H-ADAPTIVE AND P-ENRICHMENT ALGORITHMS THAT REFINE THE STRUCTURE OF THE MESH TO OPTIMIZE COMPUTATIONS. THE FULL POTENTIAL FOR USE OF SUCH SMART ALGORITHMS IN COMPUTATIONAL FLUID DYNAMICS APPEARS TO LIE IN PARALLEL COMPUTATIONS PRIMARILY BECAUSE OF THE EXTREMELY LARGE ELEMENT LIBRARIES THAT ARE COMMONLY INTRODUCED IN HP-MESH REFINEMENT SEQUENCES. A GENERAL-PURPOSE PORTABLE PROGRAMMING LANGUAGE CALLED LINDA THAT PROVIDES CONVENIENT ACCESS TO PARALLELISM ON HOMOGENEOUS NETWORKS OF COMPUTERS HAS BEEN DEVELOPED. BY USING THIS SOFTWARE, A SUPERCOMPUTING ENVIRONMENT CAN VIRTUALLY BE FORMED FROM THE UNUSED CENTRAL PROCESSING UNIT CYCLES IN AN EXISTING NETWORK. THE STRUCTURE OF THE SYSTEM PROVIDES A HIGHLY FLEXIBLE FORMAT ALLOWING SOFTWARE WRITTEN FOR OTHER PARALLEL ARCHITECTURES TO OPERATE IN THE NETWORK LINDA SYSTEM WITH MINIMAL MODIFICATIONS. IN PHASE I, A JOINT APPROACH WITH THE LINDA SYSTEM IS EXPLORING THE FEASIBILITY OF DEVELOPING PARALLEL VERSIONS OF THE NUMERICAL ALGORITHMS USED IN THE CODE. A SUCCESSFUL EFFORT WOULD PROVIDE A SIGNIFICANT BREAKTHROUGH IN PARALLEL COMPUTING.