| Passwords are one of the most important methods for identity authentication,whose security has attracted widespread attention from domestic and foreign scholars.Through the analysis and mining of real password data sets,the characteristics,distribution and reusability of user passwords are gradually discovered and understood.In this context,several password generation models have been proposed to improve the success rate of password cracking.In order to further improve the efficiency of password cracking,this paper focuses on the phases of password encryption and ciphertext matching,and proposes high-throughput implementations based on ARMv8 architecture.In the encryption phase,each round of logical computation of the hash function has strong data dependencies and cannot effectively utilize the NEON instruction set for data level parallelism.Based on the characteristics of password cracking that it is necessary to encrypt a large amount of plaintext,we propose a high-throughput multi-batch encryption scheme that can directly map to SIMD architecture and make full use of the pipeline in the modern processor architecture.The evaluation indicates that the throughput of multi-batch MD5 and SHA-1 encryption increases by 2.35 times and 41.3% respectively,compared to that of the best implementation in OpenSSL.The ciphertext matching phase uses the Bloom filter to filter out most of the absolutely incorrect ciphertext in the set,and then use the linear search method for further comparison.In this implementation,as the number of password to be cracked increases,the size of Bloom filter gradually increases,causing its memory access performance to drop dramatically.To address this problem,this paper proposes a cache-friendly two-round segmented Bloom filter implementation based on the software-managed buffer.In the experiment,for different sizes of Bloom filters,the throughput of the build phrase is increased by 1.2 to 4.5 times,and the probe phase is increased by 11% to 105%. |
PDF Full Text Request