基于MapReduce和GPU双重并行计算的云计算模型  

A Cloud Computing Mode of Dual Parallel Computing Based on MapReduce and GPU

作　　者：曾青华[1] 袁家斌[1] 

机构地区：[1]南京航空航天大学计算机科学与技术学院,南京210016

出　　处：《计算机与数字工程》2013年第3期333-336,共4页Computer & Digital Engineering

基　　金：国家高技术研究发展计划(863)(编号:2009AA044601);国家自然科学基金重点项目(编号:61139002);南京航空航天大学基本科研业务费专项科研项目(编号:NS2010230)资助

摘　　要：由于MapReduce模型进行Map和Reduce操作时需要频繁的CPU计算,面对大量并行计算任务时,CPU占用率甚至达到百分之百。而GPU有比CPU更好的并行计算能力,适度使用GPU,可降低了CPU的占用时间,又能用GPU的参与来平衡系统的计算能力。论文结合GPU技术和MapReduce技术的不同优势,设计出一种基于MapReduce和GPU双重并行计算的云计算模型。通过理论建模与实验验证,结果表明此模型可实现多GPU的MapReduce任务并行处理,提高了高性能计算的性能。The Map and Reduce operations of MapReduce model require frequentcomputing on CPU.While parallel computing tasks are massive,the CPU occupancy rate will be even up to 100%.However,GPU has the better parallel computing capability than CPU.If GPU is used appropriately,it can reduce the CPU occupancy time and balance the computing capability of the system.This paper takes the different advantages of GPU technology and MapReduce technology to build a cloud computing mode of dual parallel computing based on MapReduce and GPU.Through theoretical modeling and experimental validation,the results of this design show that this model can finish parallel processing of multi-GPU MapReduce task to improve the performance of high-performance computing.

关 键 词：云计算 MAPREDUCE模型 GPU 高性能计算 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]