Computational Electromagnetics (CEM)
Back to Applications
Principal Contact Person and Organization (including e-mail address):
Yi Pan
Department of Computer Science
The University of Dayton
300 College Park
Dayton, OH 45469-2160
pan@cps.udayton.edu
Brief Description of Application:
Computational electromagnetics (CEM) in the time domain is the most general numerical approach for describing dynamic or wide-band frequency electromagnetic phenomena. Computational simulations are derived from discretized approximations to the time-dependent Maxwell equations [1-10]. High numerical efficiency of CEM simulation procedures can be attained either by algorithmic improvements to solve the Maxwell equations or by using scalable parallel distributed memory computer systems. Since the massive volume of data processing are involved in solving the Maxwell equations, distributed memory computer systems are viable means to solve the memory shortage problem on workstations or vector computer systems. The other advantage is reduced time when parallel processing is employed to solve the Maxwell equations. Hence, parallelization of existing sequential Fortran code for solving Maxwell equations is an important effort towards developing efficient and accurate CEM code in analyzing refraction and diffraction phenomena for aircraft signature technology.
Number of Lines of Code: 1726
Target Platforms and HPF Compilers Used:
T3E (and SP2 in the future)
Coding Styles (data decompositions, computational methods):
Domain (data) decompositions
Performance Information, if Available (including any possible comparisons to MPI and/or OpenMP):
Experimental runs show that the execution time is reduced drastically through parallel
computing. Due to the requirement of the HPF compiler I used, many huge arrays have to
be initialized before their usage, a lot of overhead is introduced. However, the code
is still scalable up to 98 processors on the Cray T3E. Compared with MPI and PFA
implementations, the HPF implementation has almost the similar efficiency with much less
porting effort. Based on the experimentation carried out in this researc, we believe
that a high level parallel programming language such as the High Performance Fortran
is a fast, viable and economical approach to parallelize many existing sequential codes
which exhibit a lot of parallelism.
Please comment on any aspects of the application that might be interesting, including any problems using HPF effectively:
No so many people know how to use HPF. I cannot get help from
anyone except Mark Young of PGI. Mark Young is a wonderful
person who helped me greatly during my project.