high performance computing on graphics processing units: hgpu.org

Passwords are without doubt the most common means for authentication throughout all kinds of applications on computer systems, ranging from local or online-service user logins to the protection of sensitive data by password based encryption. However, wherever passwords are employed, these are prone to loss or disremembering, an effect which, especially driven by the advent of the internet, is even impaired by the ever increasing number of passwords an average computer user is required to manage. While many cases of password loss may be easily mitigated, e.g. by resetting an online-service password, there are other scenarios where the loss of password information will have more impact. One worst case example for such a scenario may be given by the permanent loss of important, encrypted business data following the unexpected death of the data owner. For this reason, password recovery, the lawful counterpart to password cracking, has ever since been a topic tightly related to the use of passwords and is covered by numerous software products. The efficiency of most of these applications and thus the likelihood to recover a lost password in acceptable time, traditionally has been directly linked to the available CPU processing power of a single system. Some applications like [EDPR] were able to circumvent this limitation by making use of an coordinated distributed effort across a potentially large number of available computer systems. Others even employed specialized hardware accelerators like the [TACC]1441 in order to perform a larger number of password candidate checks per time. The applicability of such solutions however is somewhat limited, as computer clusters as well as specialized hardware solutions usually are too expensive for home or small and even medium-sized business users. With modern graphic processing units, which have meanwhile in many aspects outgrown their original field of use, another, cost-efficient solution to this problem has become available. While these GPUs are still primarily used to produce display output, they now can also be employed as high-performant, massively parallel co-processing units, an application which since 2006 is even officially supported by both leading GPU vendors. As password recovery is usually a highly parallelizable task, it is an ideal candidate for GPU assistance. Thus, password recovery applications were among the early adopters of GPU processing support, which has already lead to largely improved performance for many of these applications.