Jeff Liebermann wrote:

I couldn't find any documentation as to the program limits. As I
understand it, MMANA-GAL has a stripped down NEC2 engine. 4NEC2 and
EZNEC use the full NEC2 engine from Lawrence Livermore Labs. I'm not
sure what you're building but allow me to offer a clue. I recently
threw together a model of a truly ugly tower with about 8000 segments
including about 2000 pattern lines for various grids and reflectors.
After about 14 hours running, my PIII/933MHz finally spewed out a
result (that told me that I goofed and had to make corrections). I
dragged home a P4/2.8GHz with hyperthreading, which took about 8 hours
to run. This was getting ridiculous, so I dragged the program and
model to a customers office, and ran it overnight on his P4 2.8GHz
dual core, 4GB RAM, 1333MHz FSB, SATA2 drive, etc. It doesn't get
much better. It ran in 3 hours.


If your optimization is just changing a small part of the overall
structure (e.g. the position of an antenna on the tower), then you
should be looking at the Write Greens Function and Read Greens Function
cards.
http://www.nec2.org/part_3/ngf.html

This is how they do things like analyze antenna placement on a ship or ISS.