Professional Documents
Culture Documents
Numerical Algorithms For The Hypercube Concu - 1988 - Mathematical and Computer
Numerical Algorithms For The Hypercube Concu - 1988 - Mathematical and Computer
00
Printed in Great Britain Pergamon Press plc
Jean E. Patterson
Farzin Manshadi
Rue1 H. Calalo
Paulett C. Liewer
William A. Imbriale
James R. Lyons
55
56 Proc. 6th Int. Cmf. on Mathematical Modelling
For structures being modeled by a combination TABLE 1. Timing and Speedup Factor of
of wire segments and surface patches, the the Matrix Fill for Varying
matrix equations has the form Size of Hypercube
!I :][::I=[:] t8)
Nodes
1
2
(set)
1725.0*
876.0
Factor
1.00
1.99
where a, b, c, d are sub-matrices of the
4 445.5 3.87
matrix A representing the interactions wire-
8 230.3 7.49
to-wire, surface-to-wire, wire-to-surface, and
16 122.6 14.07
surface-to-surface respectively. FN and FP
32 68.8 25.07
are the basis function amplitudes for wires
and patches, EW is the electric field at the * Timing extrapolated for l-node
center of wire'.segments, and Hp is the magnetic hypercube
field at the center of the surface patches.
Proc. 6th Int. Conf. on Mathematical Modelling 57
TABLE 2. Timing and Speedup Factor of Taflove, A., M.E. Brodwin (1975). Numerical
the Matrix Factorization for Solution of Steady-State Electromagnetic
Varying Size of Hypercube Scattering Problems Using the Time-
Dependent Maxwell's Equations. IEEE Trans.
No. of Factor Time Speedup Microwave Theory Tech., MTT-233, 623-630.
Nodes (set) Factor
Taflove, A., K.R. Umashankar (1982). A Hybrid
1 Li110.0* 1.00 Moment Method/Finite Difference Time
3141.0 1.95 Domain Approach to Electromagnetic Coup-
4' 1615.0 3.78 ling and Aperture Penetration into Complex
8 841.0 7.27 Geometries. IEEE Trans. Antennas and
16 454.0 13.46 Propagation, AP-30 617-627.
32 259.0 23.59 Umashankar, K.R., A. Tailoove (1984). Analytical
Models for Electromagnetic-Scattering,
* Time extrapolated for l-node Part II: Finite-Difference Time-Domain
hypercube Developments, Final Report in ITTRI Pro-
ject E06538, Electronics Department, ITT
Another measure of the performance is the com- Research Institute.
parison of the maximum size problem which can Yee, K.S. (1966). Numerical Solution of Initial
be executed on the hypercube versus a conven- Boundary Value Problems Involving Max-
tional sequential computer such as the VAX. well's Equations in Isotropic Media.
On the VAX 11/750, the largest finite dif- IEEE Trans. Antennas and Propagation, AP-
ference lattice which can run in a typical 14, 302-307.
user's dynamic memory allocation contains about
192,000 unit cells on the Mark III Hypercube
with 32 active nodes, the largest lattice con-
tains about 2,048,OOO unit cells.
SUMMARY
REFERENCES