A transputer supernode system is designed for studying wave scattering
in parallel. The processor farm system is enhanced by converting each
worker into a supernode comprising a transputer plus two networks of
single-processor-boards. Softwares are written to handle low-level tas
ks for parallel processing. The wave equations are solved by the metho
d of moments. The scatter is represented by a grid of 2n wires of equa
l length. Due to the manifold symmetry, the matrix can be fast enumera
ted. The matrix is also efficiently inverted by the matrix partition m
ethod. An adaptive scheme is used to evenly distribute loading among a
ll processors and minimize inter-processor communications. Significant
computational speedup is observed.