点击次数：

论文类型：期刊论文

发表时间：2015-04-01

发表刊物：Journal of Computational Information Systems

收录刊物：EI、Scopus

卷号：11

期号：7

页面范围：2615-2622

ISSN号：15539105

摘要：GPU programming model for general purpose computing is complex and difficult to be maintained. A MapReduce acceleration framework named MRCUDA is designed and implemented in this paper. There are four loosely coupled stages in MRCUDA, including Pre-Processing, Map, Group and Reduce, which can support flexible configurations for different applications. In order to take full advantage of GPU parallelism, a bitonic sorting algorithm is designed and implemented in the Group stage, and its performance is superior to general GPU sorting algorithms. Finally, according to five kinds of typical application tests, it is demonstrated that MRCUDA computing platform can reduce code scale and achieve ideal running speedup ratio. ?, 2015, Binary Information Press. All right reserved.