Distributed Password Cracking in a Hybrid Environment

distributed computing, digital forensics, cryptography, password cracking, GPGPU, BOINC

For forensic experts, encrypted data nowadays represent one of the biggest challenges. With criminal suspects unwilling to surrender their passwords, the only way to obtain the encryption key is by password disclosure or password cracking. Modern cryptographic techniques are being enhanced to provide the maximum level of security making it impossible for a single man with a computer to crack the password within a meaningful time. General-purpose computing on graphics processing units (GPUs) often accelerates the entire process. Nevertheless, a  single-machine GPU cracking has still its limits. Thus, the only way of achieving the desired amount of computational power is distributed cracking. Not in every case the forensic investigators possess a  dedicated computer cluster powerful enough to serve their needs. Complex problems may require merging multiple clusters together as well as the use of general-purpose hardware to get as much computational power as possible. The computers may be even located in geographically separated areas, e.g. multiple corporation branches, requiring a solution for secure and reliable interconnection and control of the machines. Using Berkeley Open Infrastructure for Network Computing (BOINC) framework together with our specialized modules, we designed a solution for distributed password recovery in a hybrid environment including, but not limited to, individual general-purpose CPU nodes, GPU nodes, and specialized CPU/GPU clusters. The solution is feasible for a variable number of nodes in an untrusted and unstable environment, by offering proper adaptivity and robustness. In the paper, we provide the experimental results of distributed cracking in a hybrid CPU/GPU environment. The experiments include both exhaustive search and dictionary attack. We analyze the results of different approaches using several types of hardware. Our goal is to provide a survey comparing possible attack techniques and connecting them with the most-fitting cases. The results may help forensic investigators to choose a proper method for each case.