Re: NEC-LIST: How to estimate execution time in NEC2

Abraham Rubinstein wrote:
>
> I tried to run a model with almost 2800 segments using Wiregrid, a
> graphic interface for NEC2. I selected 100 freq. steps. It gives me an
> estimated execution time of 240 hours approx. Is this true????
> ...
> How can I estimate the execution time for a model?
> Are there any key parameter to reduce this time?
> I use a PIII 500 with 256 MB RAM.

Per frequency, for large N, the time in seconds is approximately:

  8/3 * N^3 / Fps

where N is the number of segments and Fps the number of floating
point operations that are executed per second. You might think
that your PIII 500MHz can have Fps in the order of 500e6, but
for non-optimized code, this is often only 1/10th of that value.
Even then, the problem should take about 32 hours, not 240.

To get better performance, the part of the NEC code that solves
linear equations should be replaced by machine optimized Lapack
routines (or a similar math library). With that, you can come
near peak performance, which would mean about 3 hours for your
entire problem.

I believe Juergen von Hagen has posted about it earlier. If you
have kept old nec-list articles you might look it up (or is there
a repository somewhere?)

[See http://dutettq.et.tudelft.nl/~koen/Nec/ - mod.]

Greetings,
Jos

--
Dr. Jozef R. Bergervoet                      Electromagnetism and EMC
Philips Research Laboratories,             Eindhoven, The Netherlands
Building WS01                                     FAX: +31-40-2742224
E-mail: bergervo_at_natlab.research.philips.com    Phone: +31-40-2742403