RE: many segments

From: <bergervo_at_email.domain.hidden>
Date: Mon, 30 Oct 1995 13:22:11 +0100 (MET)

> Message 4/99 From DeSchweinitz DD Oct 27, 95 03:19:59 pm -0500
>
>
> Date: Fri, 27 Oct 1995 15:19:59 -0500 (CDT)
> To: nec-list_at_ee.ubc.ca
> Subject: many segments
> Sender: davem_at_ee.ubc.ca
>
> What's the hot tip for running large NEC models on a PC? I'm using the
> executable from ftp.netcom.com/pub/ra/rander/NEC, and it chokes on models
> with more that a few hundred segments or so. Is the Linux version any
> better? Is there any advantage to compiling the Fortran or c code myself?
>
> TNX
>

Just recompile with the parameters MAXSEG and MAXMAT at a larger value!
Also the parameter IRESRV should be (at least) MAXMAT**2 and your PC
should have 16*IRESRV bytes of RAM.

If RAM is insufficient, then DON'T rely on virtual memory; both the
compiler-provided and MS_Windows type are inefficient for matrix
factorization.
Instead use the NEC built-in block factorization (MAXSEG > MAXMAT). This
will, however, require 4 times the usual disk swapspace and also opening the
necessary files is omitted in some versions of the code, so you must add
those open statements.

Success..
Jos

----
    Jozef R. Bergervoet              Electromagnetism and EMC
    Philips Research Laboratories, Eindhoven, The Netherlands
    E-mail: bergervo_at_prl.philips.nl      Phone: +31-40-742403
Received on Mon Oct 30 1995 - 16:27:00 EST

This archive was generated by hypermail 2.2.0 : Sat Oct 02 2010 - 00:10:36 EDT