The new version of the program to guess passwords

There was new version John the Ripper, programs for guessing/audit of Unix-passwords (and not only Unix) on them hashes, for the first time with official support parallelization, realized by means of OpenMP directives.

At the given stage, OpenMP it is supported and effectively works for “slow” types hashes: OpenBSD-like on the basis of Blowfish (algorithm bcrypt), glibc 2.7 + SHA-crypt, Solaris SunMD5. For bcrypt the optimized code (on x86-64 calculates on two hashes in a parallel way on each stream) is used. For SHA-crypt and SunMD5 system function crypt_r (3) on glibc or a supporting multithreading crypt (3C) on Solaris (and SHA-crypt there it is supported too) is meanwhile used.

Efficiency of this approach has been checked up to release on 4 and 8-nuclear x86-64 and on UltraSPARC T2 (4 kernels, 32 streams). For bcrypt, on 4-nuclear Core i7 and UltraSPARC T2 an expedition (in comparison with one process which is not supporting parallelism) has made from 5.3 to 5.5 times (it exceeds 4 times thanking SMT). On 8-nuclear system (without SMT), the expedition has made 7.9 times. For SHA-crypt on Linux and Solaris and for SunMD5 on Solaris results are hardly worse – an expedition about 3.7 times on 4-nuclear systems. For discussed types hashes and their typical customisations (quantity of iterations which is regulated by the system administrator) speech can go about an expedition approximately with 200 to 700-1600 checked combinations {the user, the password} in a second. Further parallelism on some computers meanwhile can be carried out on-starinke (manually or with MPI).

One of the main advantages of the OpenMP-approach: simplicity in assemblage, installation and program usage. And usage differs nothing from traditional, JtR works as usually (including possibility to interrupt and continue operation from a former place), only faster. One of the main disadvantages – necessity of implementation of support for each type hashes or the interface separately.

In the future versions JtR the extension of support of parallelism and on other types hashes – with usage available in JtR the optimised code is expected.

This entry was posted on Sunday, June 20th, 2010 at 3:22 pm and is filed under News. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.