NASA
High Performance Computing
and Communications Program

Computational AeroSciences Project

Sparse Equation Solver in FORTRAN-90,C++ and JAVA

Objective: Since most structures and many electromagnetic matrices are sparse, dense, skyline or banded, equation solution techniques are impractical. Thus, NASA Langley's all-FORTRAN vector sparse solver, VSS, is used to solve such applications at triple the speed of the best commercial sparse solver using assembly code (measured by SGI-Cray). FORTRAN-90, C++ and JAVA are growing in importance, and it is natural to extend VSS to exploit their portability, growing user community and easy network access.

Approach: VSS was converted to C++ (VSSC) and to JAVA (VSSJ). Simultaneously, a FORTRAN-90 version VSSse combining structures (REAL) and electromagnetics (COMPLEX) data types was developed.

Accomplishment: The performance of the VSS equation solver has been preserved in a unified structures-electromagnetics sparse equation solver,VSSse. The C++ and JAVA versions permit hardware independence and remote equation solution via Netscape, a concept that was unthinkable by many even a year ago. Performance of VSSJ (see figure) should improve markedly as JAVA matures and VSSJ code is optimized.

Significance: As technology evolves away from high-cost specialized computers toward powerful network servers and workstations (with high-performance JAVA), distributed equation solution using remote data may become the preferred solution strategy of the future. The NASA FORTRAN-90, C++ and JAVA equation solvers are on the cutrting edge of this new technology. They are being widely adopted by industry.

Status: http://transit.larc.nasa.gov/www-csb/ESP.html describes access to VSSJ and other versions of VSS.

Contact:

Dr. Olaf O. Storaasli

NASA Langley

O.O.Storaasli@larc.nasa.gov

757-864-2927