Parallel Implementation And Optimization Of Hash Algorithm Based On GPU

With the rise of GPU mode of high-performance computing, the advantages of GPU computing model gradually be reflected, the vast scale of its application has also obtained a corresponding degree of expansion. The other hand, all password protection technology continues developing, mainly the one-way hash algorithms'complexity increasing, leads to a large computation and difficulty of the search. For the traditional CPU hash algorithm, the password recover process is a process of iteration, with low efficiency and low speed calculation. Therefore, the use of traditional methods of CPU is unable to meet the needs of today's applications, there is an urgent need for a new technology with more computing power to solve the problem.This article focus on seven one-way hash algorithms including MD5, NTLM, MSCACHE, MYSQL5, DOMINO, UNIX_DES and UNIX_MD5,first analyzes principles of these hash algorithms,and then uses massive parallel computing power of the GPU to realize GPU parallel algorithm version. Specific approach is: the appropriate cryptographic hash algorithms package corresponding GPU external interface, call GPU external interface in the CPU.GPU, through its internal thread execution Manager ,controls hash generater and generate the value, it produces more than one possible different hashed values. These, it produces many different hash value after you enter the different threads of the GPU, these threads can proceed parallel computing for different passwords, compares with the known hash value, and final output of the comparison is obtained. The number of GPU threads are usually up to 100,000, so parallel computation of cracking process have very high efficiency.In order to improve the speedup of the whole program, this article presents a series of hash algorithms optimization, the main optimization strategies are: parallelization strategies setting, texture memory , constant storage optimization, global memory coalesced, CPU and GPU parallel optimization discrete character space optimization,and so on.The experiment result shows these strategies achieve desired purpose.