Electron Optical System Cad Software For High Performance Computing Research

Micro wave Tube CAD software need to deal with huge number of data and compute for a long time when it simulates a whole 3D numerical problem. So the performance is very important to the software. And if need to improve the performance of program, we need to know about the technology of profiling program,normal knowledge of improving program, well using of RAM and cache.As we know that we should make sure what need to improve, bur not improve that is not important to the program. If the program block only runs a little time, it doesn't worth to improve, cause even if it doesn't run after your improving, the whole could only save a little time. So we need to profile the program to detect where is the key part of performance. We could use some tools and measure to detect the hotspot of the program. As have found the hotspot, we have a definite object in view. In this study we both read the program and analyze the run time of it, and then use the performance analyzing tools of Microsoft Visual Studio 2005 to detect the hotspot. And we found that the arithmetic of search electron in the finite-element module takes the most time of the program. There are both scope-first and deepness-first ways of searching, and named scan arithmetic and pursue arithmetic in this study. And at the same time it shows the assistant arithmetic.This study also improve other important part of EOS(Electron Optics Simulator), just as the function of calculating electric field, and used some technology of improving program. And then do some work to save RAM space and use RAM more quickly. Then the study compare the performance of unimproved and improved program used performance analyzing tools of Microsoft Visual Studio 2005. And used both unimproved and improved EOS to compute a real module and compare the real run time again.Nowadays multi-core PC have been used to every area such as daily life and science study, if Micro Wave Tube CAD could run on a multi-cell computer, and parallel run on multi-core, its performance could be markedly improved. So this study also discuss about parallel programming, introduce about parallel computer system, parallel programming module, parallel arithmetic, parallel programming technology and how to value the performance of parallel program, and compare several parallel programming circumstance such as MPI, HPF and OpenMP.As introduced in the study OpenMP is the most suitable to parallel program on multi-core computer. It show s simple example that using OpenMP to parallel program on multi-core, and show its performance. Then try to use OpenMP in the part of calculating trace in EOS, and also test its paralleling performance. Conclude and study valuable experience of parallel programming, and definitude the way we should go to parallel program in EOS.