An experimental study of BSP sorting algorithms

The bulk-synchronous parallel (BSP) model of computation has been proposed as a unifying and bridging model for the design, analysis and implementation of parallel algorithms that are both scalable and portable among diverse parallel machines. The authors implement, using the Oxford BSP Toolset, BSPlib, a number of BSP deterministic and randomized sorting algorithms on a variety of parallel systems, including the SGI Power Challenge and the CRAY T3D. The experimental results illustrate the efficiency, portability and reusability of the algorithms across such diverse hardware platforms, and reinforce their claim that the BSP model can be used as a kind of programming paradigm.