Block-cyclic array redistribution on networks of workstations
Résumé
This article deals with the run-time comparison between several algorithms (including the MPI_Alltoallv() function call) to perform the redistribution of arrays that are distributed in a block-cyclic fashion over a multidimensional processor grid. The generation of the communication messages to be exchanged by the processors involved in the redistribution is not taken into account. Rather we focus on the scheduling of those messages: how to organize the message exchanges into "structured" communication steps that will minimize communication overhead.