Parallel matrix multiplication on networked microcomputers (Q1207446)

The authors present methods for utilising parallel processing capability of the idle microcomputers on a local area network to perform computationally intensive operations encountered in linear algebra. This is achieved by programming the local area network to emulate a shared- memory multiprocessor, where data files on the file server take the role of the shared main memory. As an example, techniques for efficient implementation of level 3 BLAS (matrix-matrix operations) primitives on networked microcomputers are considered. The numerical results presented indicate that the efficiency of the parallel algorithm is highly sensitive to the delay factor (the smallest amount of time between the starting times of any two nodes) and also the block size.