| With the increasing of national economy, analysising electromagnetic compability(EMC) of big size platform is put in an important position. This thesis utilize MOM to solve the electromagnetic integral equation in which the key point is use LU factorization to deal with the big size matrix equation. The method of LU factorization not only is slow in speed but also needs huge of memory, so it is necessary that using parallel computing enhance the speed and letting out-of-core satisfy little memory. The main job of this thesis is programming both parallel in-core MOM and parallel out-of-core MOM, and then comparing their characters with numerical testing to some different examples. Thus in this way it is evaluated if the parallel out-of-core MOM can be applied for the big size platform.At the first, chapter 2 introduces the parallel computing which consists of hardware, software, parallel LU factorization and parallel out-of-core technology. Generally, parallel computing needs large server for its hardware and MPI parallel package for its basic software environment. When proceeding LU factorization, it must use Sca LAPCK which includes BLACS, LAPACK, Blas, Pblas and so on. There is function Px GETRE for LU factorization of in-core version and out-of-core version, which utilizes block cycle for matrix storage and left-looking LU factorization.In the chapter 3 MOM basic theory is introduced. For using MOM model is splitted to triangles in size of one point ten wavelength, and then a matrix equation is got by spreading the electromagnetic field integral equation based on RWG basis function which is solved by LU factorization. While in chapter 4 some methods is introduced for the important singular integral problem when getting the EFIE and MFIE by MOM. The singularity of EFIE can be easily handled by classical method, however, the singularity of MFIE is hard to deal with, as may cause the accuracy of MFIE is not better than that of EFIE.At the last, chapter 5 illustrates some testing examples of three different models which are sphere, missile and F22. The testing results shows that both of in-core MOM and out-of-core MOM are accurate with the testing size of block being 128 and proper process grid being square. It is noticed that memory need of out-of-core MOM is much less than that used for in-core MOM, while computing speed of the former is a little slow than that of the latter. So out-of-core MOM is very good in performance, and the scale of current computing can be rewritten in the server. The only weakness of this out-of-core MOM is when filling the matrix to hard disk too much time is needed, so it could be the next target. |
