PowerBlade wrote:
Did you try to do an implementation of it or is it just theoretic so far?
Did you do some math of how many merging chains you get with only using 32 bits compared to the 64 bit rainbow crack use?
post5758.html#p5758I implemented it but it was a very specific case password lengths 1-6 and used reversing. The 32 bit version on a 2.5 GHz Core2Quad gets 87.6 million links/sec. I rewrote this to use CUDA, but I hard coded in the Omni-6 through Omni-8 key spaces.
It should be a little worst for chain merges than the current method even though I'm using 128 bits form the hash instead of 64 bits like rcrack. So LM (and variants), MySQL323, and Cisco PIX won't work (without modifications which will/might cause problems). LM could be done with all-space#7-7 nicely if you take 3 characters from each 32 bit integer and then 23 and 3 from each and multiply then the "3" by 23 and add them. MySQL323 (62 bit hash) rainbow tables are pointless anyway. Cisco PIX (96 bit hash) which will cause problems if you want to do more than one key space block like numeric#1-7 which would be 7 key space blocks. Right this method doesn't work well if you do 1-n. You need to do a minimum of 5 (5-n). So that the smallest key space block is not negligible compared to the total key space. With Omni-6 the smallest key space block is 1/49,973 the total key space and 1/348,168 for Omni-7. Since you can just brute force 0-4 in under a second it doesn't really matter.
For a character set length of 95 you'll probably get bad results if you take more than 3 characters from a 32 bit number which limits you to 9 characters which is 2^59.13. You can do a 10 character long password with 95 characters (2^65.70) in a single rainbow table set if you separate the key space into different blocks such as [ -\+][ -~]{9}, [,-7][ -~]{9}, [8-C][ -~]{9}, [D-O][ -~]{9}, [P-\[][ -~]{9}, [\\-g][ -~]{9}, [h-s][ -~]{9}, and [t-~][ -~]{9}. Which is just like having 8 hybrid key spaces blocks combined which is allowed with my method.