Matrix Inversion in Single and Multi-processor Environments

The inversion of matrices was calculated on a single transputer and on a network of transputers to show the advantages of parallel processing. The adjoint-determinant method of calculating the inverse was used. The software developed for the single transputer configuration was only slightly modified to execute on the multi-transputer network. The Transputer Development system (TDS) was used for all development, compiling, and execution. The OCCAM language was used for software development. This development system and the application software executed on INMOS B003 and B004 evaluation boards with an AT&T 6300 computer used as a host. The tests executed showed the single processor configuration was slower than the multi-processor configuration. The difference in the single and multiprocessor processing times increased as the matrix size was increased. It was determined the communications overhead resulting from the parallel processing did affect the total processing time in some instances. The accuracy associated with the inversion of several ill-conditioned matrices was about the same under both environments. These results lead to the conclusion that parallel processing is beneficial for large problems where communications overhead is a small percentage of the total processing time