Clustering algorithm optimized for Cell Broadband Engine Architecture

Abstract

In this paper, we want to evaluate the performance of Cell B.E. processor which is based on the Cell Broadband Engine Architecture (CBEA). For this purpose, we chose a clustering algorithm that we have optimized for this architecture by efficiently harnessing the facilities provided. Performance of the Cell B.E. processor was evaluated by executing the algorithm using computation in single and double precision. In both cases, performance was evaluated with and without SIMDization. For single precision, we obtained a maximum speedup of 29.07 by activating 6 SPE processors without SIMDization and a speedup of 10.9 for 6 SPE processors with SIMDization. For double precision, we obtained a maximum speedup of 14.51 by activating 6 SPE processors without SIMDization and a speedup of 8.34 for 6 SPE processors with SIMDization.

Keywords: Cell Broadband Engine Architecture, DMA, SIMD, speed-up, parallel computing