Matrix multiplication on a multicomputer system with switchable main memory modules

The issues involved in implementing algorithms for large-scale numeric computations, specifically matrix multiplication, in a dynamically partitionable multicomputer system (SM3) are discussed. SM3, a multicomputer system with switchable main memory modules, supports the efficient execution of parallel algorithm by (1) allowing the sharing of switchable main memory modules between computers, (2) supporting network partitioning, and (3) employing global control lines to efficiently support inter-processor communication. Timing equations are derived for the matrix multiplication operation and an analysis is presented. The SM3 is compared with a conventional network system in order to highlight the differences in the two architectures. 13 refs., 8 figs.