Distributed Rainbow Table Generating And Search Based On Checkpoint

Hash function can convert different messages into fixed-length hash values.Without hash collision,different messages have different hash values,so hash function is widely used in password verification and other fields.Rainbow table is a hash function crack method based on time-space tradeoff,it's the most effective crack method for hash function without salt.In recent years,heterogeneous parallel computing develops rapidly.Intel has introduced MIC coprocessor,and GPU has also been widely used in general-purpose computing.Rainbow table contains three processes: offline computation,online computation and chain regeneration.The computation of rainbow table can be highly parallelized,it's very suitable for MIC or GPU.In this paper,we designed and implemented a rainbow table system for cracking large batches of hash values,it can run on multi-node multi-card platforms.The system contains three modules: rainbow table generation,original data reorganization and rainbow table search.The rainbow table generation module can run on TH-2 supercomputer,and accelerate offline computation via MIC coprocessor.The rainbow table search module runs on multi-node,nulti-GPU clusters,and accelerates computation via GPU,this module makes GPU computation,CPU computation,network and disk I/O parallelized and pipelined,can take full advantage of the components of the platform.And we also applied checkpoint in this system,to reduce false-alarm rate,and made the large batches of hash value cracking more efficient.In this paper,we did lots of work on the parallel design and performance optimization based on the hardware characteristics of the experimental platform.According to the experiment results,the system is ideal for both crack success rate and crack speed.