Torus data flow for parallel computation of missized matrix problems

AbstractA data-flow approach is used to solve dense symmetric systems of equations on a torus-connected 2-D mesh of processors. A torus mapping of the matrix onto this processor array allows the Cholesky decomposition to be completed in 3n − 2 time steps using only n2/4 processors (less than half the number needed in previously reported results). New definitions for missized problems and parallel algorithm performance are given along with various time-step, efficiency, and processor utilization plots