Attn: Tina Mihaly Pauna Editor, CSCC Update Technical Editor, Caltech CCSF 3312 Dunbar Ave. Ft. Collins, CO 80526 From: Gang Cheng, NPAC, Syracuse University Dear Tina: I have enclosed three hardcopies of the same color image for you to choose. Please let me know if you have any problem with this submission. Thanks. (submitted to CSCC (Concurrent Supercomputing Consortium) annual report) Parallel Implementation of a MoM-based Computational Electromagnetics Application on Intel Machines Gang Cheng (gcheng@npac.syr.edu)*, Kenneth A. Hawick (hawick@npac.syr.edu)*, Jay Mortensen (jay@midas.syrres.com)**, Geoffrey C. Fox (gcf@npac.syr.edu)* and Chung-chi Cha (cha@syrres.com)** * Northeast Parallel Architectures Center at Syracuse University ** Syracuse Research Corporation Corresponding Author: Gang Cheng, gcheng@npac.syr.edu, (315)443-2083, (315)443-1973 (fax) Topic Area: Computational Electromagnetics A group of computational scientists teamed at Northeast Parallel Architectures Center (NPAC) and Syracuse Research Corporation (SRC) recently completed a project to port an industrial computational electromagnetics (CEM) simulation code to a range of high performance computing platforms using the SPMD programming model. Two of the most powerful platforms we were able to exploit were the Intel Delta and Paragon computers systems at Caltech. The ParaMoM method-of-moments CEM code requires the construction, factorization, solving and subsequent decomposition of very large dense complex matrices. To achieve a portable scalable code we employed the ScaLAPACK linear algebra package from the Univerity of Tennessee (and which is available through the Center for Research on Parallel Computation's National Software Exchange project). We implemented a complex version from an early release of the single precision ScaLAPACK package. We also used Tennessee's BLACS communications software package which is implemented on the NX/2 message passing system provided by Intel on both the Delta and Paragon machines, but is also available on other message passing systems to allow us to port our completed application code to systems such as clusters of DEC Alpha workstations, IBM RS/6000 clusters, SUN workstation clusters and also to the IBM SP1 system. The facilities supplied to us through the CRPC and the CSCC have been invaluable to us in testing and evaluating our ported code. In particular, we were able to achieve long test and production-length runs on the Delta and Paragon systems, that would not have been possible elsewhere. The color figure shows the resulting electromagnetic far-field that we obtain from a realistic airborne system in calculating its radar cross section. It is important to be able to simulate the radar cross section of aircraft accurately as part of the design optimization process. The ParaMoM code has significant advantages over simpler direct method codes in this area, as it allows stable and accurate results on three dimensional systems where the curvature parameterization follows arbitrary surface geometries exactly. The code is computationally intensive and due to the large number of basis functions required for simulating realistic geometries such as the aircraft system shown, it also requires very large memory on the computer system. Our ported code is able to exploit the large amounts of distributed memory available on the Delta and Paragon systems to best effect. This project was carried out by the Northeast Parallel Architectures Center for Syracuse Research Corporation under the InfoMall technology transfer program. The project achieved production of a portable, scalable, accurate and maintainable radar cross section code that has cost effectiveness benefits for SRC and its customers. Detailed information about this project can be found on NPAC's Mosaic Server. URL: http://www.npac.syr.edu/NPAC1/PUB/gcheng/CEM/home.html. Caption of the color image --- "Far-field Electromagnetic Scattering from an aircraft, computed by the ParaMoM-MPP code on an Intel Paragon"