本文選題：大數(shù)據(jù)處理 + 異構(gòu)計(jì)算　； 參考：《實(shí)驗(yàn)室研究與探索》2017年01期

【摘要】：在大數(shù)據(jù)時(shí)代,兼顧大數(shù)據(jù)處理與高性能計(jì)算是目前對(duì)計(jì)算機(jī)系統(tǒng)的迫切需求。針對(duì)Spark大數(shù)據(jù)處理與基于GPU的高性能計(jì)算,分析了基于GPU的Spark技術(shù)。它主要通過構(gòu)建CPU和GPU的異構(gòu)并行,使計(jì)算機(jī)獲得強(qiáng)大的計(jì)算能力,并在實(shí)驗(yàn)室環(huán)境下探討了Spark-GPU技術(shù)的實(shí)現(xiàn),闡述了算法實(shí)現(xiàn)的技術(shù)流程。在此基礎(chǔ)上,通過仿真實(shí)驗(yàn)評(píng)估了Spark和Spark-GPU技術(shù)的性能。實(shí)驗(yàn)表明,Spark-GPU技術(shù)可以達(dá)到上百倍的加速比,這對(duì)圖像處理以及信息檢索等領(lǐng)域的發(fā)展都具有重要推動(dòng)作用。
[Abstract]:In the era of big data, it is an urgent need for computer system to combine big data processing with high performance computing. Aiming at Spark big data processing and high performance computing based on GPU, the Spark technology based on GPU is analyzed. By constructing the heterogeneous parallelism of CPU and GPU, the computer can obtain powerful computing power. The realization of Spark-GPU technology is discussed in the laboratory environment, and the technical flow of algorithm realization is expounded. On this basis, the performance of Spark and Spark-GPU technology is evaluated by simulation experiments. Experiments show that the Spark-GPU technology can achieve a speedup ratio of several hundred times, which plays an important role in the development of image processing and information retrieval.
【作者單位】： 南京信息工程大學(xué)電子與信息工程學(xué)院;
【基金】：國(guó)家自然科學(xué)基金(NSFC61203273) 江蘇省自然科學(xué)基金(BK20141004)
【分類號(hào)】：TP311.13
，

本文編號(hào)：1939755