An Experimental Implementation of Migration Algorithms On the Intel Hypercube
AbstractThis study was aimed at testing whether the Intel Hypercube is suitable for geo physical data processing—i.e., whether the multiprocessing capability of the Hy percube can achieve high performance at low cost in carrying out large-scale geo physical data processing tasks. Two basic migration algorithms—phase shift plus in terpolation migration, and second order in time, fourth order in space, finite differ ence migration—were implemented on the Hypercube. The multiprocessing algo rithm designs were based on the fol lowing issues: control, partitioning, synchronization, scheduling, communica tion, and optimization. The concurrent pro cessing in the d6 Hypercube speeds up the execution by three to four fold over a VAX 11-780. However, the limitations on the size of array by the available memory in the cube manager and the nodes, and the overhead incurred for communication and synchronization, make the present Hypercube system unsuitable for realistic geophysical data processing.