Design Of A High-performance Password Cracking Platform Based On MPI And GPU Computing

As an effective means of protecting personal information and identification, password is used widely, so that its safety has increasingly received more attention. Password brute force is an important research direction of cryptography, and also one of the most effective methods for the security, intelligence, military and other departments to obtain all kinds of information. However, the feasibility of brute force is always limited by its long cracking time. With the rapid development of high-performance computing, distributed GPU cluster technology to crack the password has made many achievements, followed by a crack application scalability problem, which means hard to support the platform in addition to other hack functions. The scalability problem requires a separate study and deployment for each password-cracking.Distributed computing and GPU parallel computing are effective methods to increase computing speed. Because the platform this thesis studies does not involve large-scale file storage or data mining, but mainly involve in data communication among computing nodes, and MPI with such features as flexible support for parallel algorithm implementation, high communication efficiency and good portability, this article select MPI as a distributed application tools, rather than Hadoop or Spark. Focusing on how to improve the efficiency of password cracking and solving the scalability problem, this thesis studies a high-performance password cracking platform based on MPI and GPU computing, the main content of this thesis can be summarized as follows.(1) A three-tier architecture model, includes the application layer, communication controller layer, and GPU computing layer, is proposed to solve the problem of scalability. And the interfaces between each two layers are designed. In the basis of the model and the interfaces above, the programs of the application layer and communication controller layer are developed;(2) A distributed password cracking platform architecture is designed to run the crack function distributively, and the deployment processes of the platform are elaborated.(3) That the problem of MPI and GPU could not work together caused by Windows Display Driver Model is solved, and the problem caused by Timeout Detection and Recovery of GPUs, which leads to GPU program access time limited, is solved as well.(4) Experimental tests are carried out to prove that the three-tier architecture model and the distributed password cracking platform architecture does work. The tests show that the platform can effectively reduce the password cracking time. However, due to password’s uneven distribution in dictionary concentration, the password cracking time and the number of the cluster hosts are not linearly related.