基于GPU-Hadoop的并行计算框架研究与实现

Study of parallel computing framework based on GPU-Hadoop

针对原生的Hadoop云平台处理海洋环境信息可视化效率不高的问题,提出了一种GPU嵌入Hadoop云平台的并行计算框架。该框架以原生Hadoop为基础,GPU并行计算与MapReduce相结合,实现了高效的海洋流场可视化和特征可视化。实验结果表明,提出的并行计算框架在处理数据密集型和计算密集型的海洋数据的效率上优于原生的Hadoop云平台,可达到6～8倍的加速比。因此,提出的云平台框架可以有效提高海洋信息可视化的计算效率,对我国海洋事业的信息可视化发展具有重要的推动作用。

This paper proposed an parallel computing framework based on Hadoop embedded GPU for improving the efficiency of ocean data visualization. Fistly,this framework was based on the original Hadoop. Then,it combined by GPU parallel computing and MapReduce parallel processing mechanism. Finally,this framework achieves high-efficiency flow visualization and feature visualization. Experimental results show the proposed framework achieved higher efficiency than the original Hadoop,the speedup rate reaches six to eight. Therefore,the proposed framework plays a very important role in improving computing efficiency and developing of ocean information visualization.