In many practical applications, the electrically large problem contains the fine structures, so the problem will get much larger in terms of cells or unknowns since the mesh must change gradually from the smaller cell size to larger ones. Any jump of the adjacent cell sizes will introduce significant numerical reflection from this jump. A reflector antenna with a circular horn feed is a good example to demonstrate the issue described above. For example, a PEC reflector antenna is not a complex problem if the feed is not present, which can be easily simulated by most of the existing software packages. Since the feed is involved in problem, the reflector system becomes a very difficult one for most of the software packages. If we handle the feed and reflector separately, we have to ignore the higher order coupling between the feed and reflector, and this will introduce a significant error into the side lobe of the antenna far field pattern. Because GEMS can use multiple processors to simulate the problem, so it is not a big problem to GEMS at all.