1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
|
<!DOCTYPE HTML SYSTEM "html.dtd">
<HTML>
<HEAD>
<TITLE>Execution Time</TITLE>
</HEAD>
<BODY>
<CENTER>
Execution Time
</CENTER>
<P><A HREF="../exetime.html">Measured times</A> on various platforms </P>
<P>The program execution time depends on the number of patches and the number
of wire segments used. The central processor time approximately follows the
formula; </P>
<DL>
<DT>T=</DT>
<DD>T<SUB>1</SUB> + T<SUB>2</SUB> + T<SUB>3</SUB>+ T<SUB>4</SUB> </DD>
<DD>T<SUB>1</SUB>=(A<SUB>1</SUB>kN<SUB>s</SUB><SUP>2</SUP> +
A<SUB>2</SUB>kN<SUB>s</SUB><SUP>2</SUP> +
A<SUB>3</SUB>kN<SUB>s</SUB>N<SUB>p</SUB> + A<SUB>4</SUB>N<SUB>c</SUB>)/M, </DD>
<DD>T<SUB>2</SUB>=B(N<SUB>s</SUB> + 2N<SUB>p</SUB>)<SUP>3</SUP>/M<SUB>2</SUB>,
</DD>
<DD>T<SUB>3</SUB>=CN<SUB>s</SUB> +2N<SUB>p</SUB>)<SUP>2</SUP>/M , </DD>
<DD>T<SUB>4</SUB>=DkN<SUB>f</SUB>(N<SUB>s</SUB>+2N<SUB>p</SUB>), </DD>
<DT> where </DT>
<DD>N<SUB>s</SUB>=number of wire segments, </DD>
<DD>N<SUB>p</SUB>=number of surface Patches, </DD>
<DD>N<SUB>c</SUB>=number of connections between a wire and surface, </DD>
<DD>N<SUB>e</SUB>=number of different excitations, </DD>
<DD>N<SUB>f</SUB>=number of far-field calculation points </DD>
<DD>M=number of degrees of symmetry, </DD>
<DT>k=</DT>
<DD>1 for structure in free space, </DD>
<DD>2 for perfect ground of reflection coefficient approximation, and </DD>
<DD>4 for Sommerfeld/Norton method. T<SUB>1</SUB> is the time to fill the
interaction matrix; T<SUB>2</SUB> is the time to factor the matrix;
T<SUB>3</SUB> is the time to solve for the currents for all excitations; and
T<SUB>4</SUB> is the time to calculate far fields. <P>The proportionality
factors depend on the computer system on which the program is run. The factors
in seconds for a CDC 7600 computer when the matrix fits in core are roughly
</P>
</DD>
<DD>A<SUB>1</SUB>=3.(10<SUP>-4</SUP>), </DD>
<DD>A<SUB>2</SUB>=5.(10<SUP>-5</SUP>), </DD>
<DD>A<SUB>3</SUB>=5.(10<SUP>-4</SUP>), </DD>
<DD>A<SUB>4</SUB>=2.(10<SUP>-2</SUP>), </DD>
<DD>B=2.(10<SUP>-6</SUP>), </DD>
<DD>C=4.(10<SUP>-6</SUP>), and </DD>
<DD>D=6.(10<SUP>-5</SUP>), When the extended thin-wire kernel is used,
A<SUB>1</SUB> is increased by about 18 percent. If the approximation for large
interaction distances is used with RKH=R<SUB>o</SUB>, then A<SUB>1</SUB> is
multiplied by (1 - 0.7F) where F is the faction of all segment pairs for which
the separation is grater than R<SUB>o</SUB>. <P>Unless a large number of
excitations or far fields are requested, T<SUB>1</SUB> and T<SUB>2</SUB> will
account for nearly all of the running time. If the matrix does not fit in core
storage, T<SUB>1</SUB> and T<SUB>2</SUB> will be larger than indicated above.
They may be much larger if I/O time is included. </P>
<P>The code SOMNEC requires about 15 sec to write the Summerfeld/Norton data
file on a CDC 7600 computer. </P>
</DD>
</DL>
<HR>
</BODY>
</HTML>
|
