Talk:MD5CRK

{{WikiProject banner shell|class=Start|

{{WikiProject Computing|importance=}}

}}

The "perspective" section is wrong. An MD5 operation takes longer a floating point operation.

216.27.176.198 09:29, 30 July 2007 (UTC)

Link to "Virginia Tech's System X" is broken (for me at least). ---- 131.216.134.91 (talk) 19:03, 16 November 2007 (UTC)

There are a number of odd things in this article: Almost certainly no cycle finding algorithm was used, since the search was distributed. I.e. one would start at a random value, repeatedly hash that value until a distinguished point is found and collect starting point and end point of these hash chains. Once a collision in the end points is found, one likely has a collision somewhere in the hash chain and hence two colliding hash values, but no cycle. The math part is sloppy too. The first formula appears to be the probability of not finding a collision after K hash computations and not as claimed the expected time to find a collision. As noted by User:216.27.176.198 the perspective underestimates the time it would take to find a collision. E.g. the Crypto++ benchmarks achieve about $2^\{23\}$ MD5 computations per second on a 1.83 GHz Intel Core 2 using one core. Using this benchmark, I get about 80'000 CPU-years to find a collision. That is of course feasible, but quite a bit longer than what the article suggests now. From archieved web pages it appears that less than 1% of the expected computation was finished in about 6 months while the project was running. Maybe the page should just be deleted, because the project did not achieve a notable result. 85.2.34.224 (talk) 10:46, 17 November 2007 (UTC)