GPU 异构系统上针对 PowerPoint的彩虹表攻击实现

摘要

In this paper , we designed and implemented rainbow tables , which is known as the most ef-ficient time-memory tradeoff method , to crack PowerPoint Password on GPU -based heterogeneous system.On pre-computing phase , We generated rainbow tables through two ways:OpenMP parallel model and CUDA parallel model .And result reveals that OpenMP is more efficient .On online phase of cryptanalysis , we gave a complete analysis of the effect of multiple checkpoints on reducing the cost of false alarms , and took advantage of it for load balancing between CPU and GPU .The average time of a-nalysis can be reduced by 75%comparing with native GPU .%彩虹表技术是当前口令恢复最有效的手段。本文使用彩虹表技术在CPU-GPU异构平台上实现对PowerPoint文件的口令恢复，在彩虹表生成阶段采用OpenMP和CUDA两种多线程模型协同工作，实验结果表明OpenMP多线程模型更适合于表生成。在彩虹表在线分析阶段，通过将再生成彩虹链移至CPU的方式减少CUDA模型束同步串行机制对效率的影响，与此同时引入检查点减少假警带来的多余工作量，最终使在线分析效率提高75％。