基于CTR模式的GPU并行AES算法的研究与实现  

Research and Implementation of GPU Parallel AES Algorithm Based on CTR Model

作　　者：费雄伟[1] 李肯立[1,2] 阳王东[1,2] 

机构地区：[1]湖南城市学院信息科学与工程学院,湖南益阳413000 [2]湖南大学信息科学与工程学院,长沙410008

出　　处：《小型微型计算机系统》2015年第3期529-533,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金重点项目(61133005)资助;国家自然科学基金项目(90715029;61070057;60603053)资助;益阳市科技计划项目(2014JZ37)资助

摘　　要：为了对任意长的明/密文进行并行加密/解密,在分析了AES(Advanced Encryption Standard)的多种模式后,采用了最合适的CTR(Counter)模式设计和实现了GPU(Graphics Processing Unit)并行AES算法,并进行了优化.针对Nvidia Geforce GTX460平台理论分析了CTR模式的并行AES算法和串行AES算法的时间复杂度,得到综合的加速比为31.59,然后在Nvidia Geforce GTX 460平台上实验运行,结果显示CTR模式的AES-256的GPU并行算法相对串行CTR模式的AES算法,实验加速比跟理论加速比基本吻合.在此基础上,对CTR模式的AES-256进行了优化.实验结果显示,优化的CTR模式的AES-256并行算法在加速比上随着明文的增大提升的比例渐渐减少并趋于稳定.优化的CTR模式的AES算法加密数据量小的明文时,其优化效果更为明显,故能有效地提升SSL(Secure Socket Layer,其明文区间为35KB-150KB)的加密性能.In order to concurrently encrypt plain text or decrypt cipher text with arbitrary length, firstly analyze various models of AES （ Advanced Encryption Standard）, then design and implement GPU （Graphics Processing Unit） parallel AES algorithm adopting the most suitable CTR （ Counter ） model, at last optimize this algorithm. The time complexity of parallel AES algorithm on Nvidia Geforce GTX 460 platform and the time complexity of Serial AES algorithm were analyzed,from which synthetical speedup ratio of 31.59 was calculated. Some experiments were executed on Nvidia Geforce GTX 460 platform and their results show that the speedup ratio of AES-256 GPU parallel algorithm compared serial AES-256 algorithm basically coincides the speedup ratio on theory. AES-256 based on CTR model was optimized on this base. The result of experiments shows that the speedup ration of the optimized AES-256 parallel algorithm gradually decreases as plaint text size increasing and verges to stabilization. When the optimized GPU parallel AES algorithm encrypts the plain text with small data size,its optimization efficiency is more obvious. Therefore it can efficiently improve the encryp- tion performance of SSL（ Secure Socket Layer） ,whose plain text has the size varying from 35KB to 150KB.

关 键 词：计数器模式 GPU并行 时间复杂度 加速比 统一计算架构 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]