Efficient hybrid parallelisation of tiled algorithms on SMP clusters

摘要

This paper emphasises on load balancing issues associated with hybrid parallelisation of tiled algorithms onto SMP clusters. The intrinsic load imbalance in hybrid parallelisation derives from the fact that message passing libraries often provide limited multi-threading support, thus allowing only the master thread to perform inter-node message passing communication. In order to mitigate this effect, we propose a generic method for the application of load balancing on the coarse-grain hybrid model for the appropriate load distribution among the working threads. We investigate both static and dynamic load balancing, and experimentally evaluate three balancing variations against kernel benchmarks.