Execution Time

Measured times on various platforms

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;

T1 + T2 + T3+ T4
T1=(A1kNs2 + A2kNs2 + A3kNsNp + A4Nc)/M,
T2=B(Ns + 2Np)3/M2,
T3=CNs +2Np)2/M ,
Ns=number of wire segments,
Np=number of surface Patches,
Nc=number of connections between a wire and surface,
Ne=number of different excitations,
Nf=number of far-field calculation points
M=number of degrees of symmetry,
1 for structure in free space,
2 for perfect ground of reflection coefficient approximation, and
4 for Sommerfeld/Norton method. T1 is the time to fill the interaction matrix; T2 is the time to factor the matrix; T3 is the time to solve for the currents for all excitations; and T4 is the time to calculate far fields.

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

C=4.(10-6), and
D=6.(10-5), When the extended thin-wire kernel is used, A1 is increased by about 18 percent. If the approximation for large interaction distances is used with RKH=Ro, then A1 is multiplied by (1 - 0.7F) where F is the faction of all segment pairs for which the separation is grater than Ro.

Unless a large number of excitations or far fields are requested, T1 and T2 will account for nearly all of the running time. If the matrix does not fit in core storage, T1 and T2 will be larger than indicated above. They may be much larger if I/O time is included.

The code SOMNEC requires about 15 sec to write the Summerfeld/Norton data file on a CDC 7600 computer.