Feature requests?

[Bitweasil]

I'm recoding the RT stuff. Features on the list:

- Multi-GPU cracking
- Multi-hash cracking (give it a hash list)
- OS X/Windows support (Cracker first, then generators)
- General code cleanups

Any other feature requests?

[kevinkk5]

a greater chainlength would be nice.

[Bitweasil]

The software supports it... It just drives cracking times up. What were you thinking?

[kevinkk5]

1,000,000,000 would be interesting. but would take too long. recently i chose a too high chainlength so i got an error but i do not know which i took. and i´m not in linux at the moment

[Bitweasil]

1,000,000,000 would be interesting. but would take too long. recently i chose a too high chainlength so i got an error but i do not know which i took. and i´m not in linux at the moment

No, that wouldn't. That would be stupid.

I believe I may have hardcoded 1,000,000 as an upper limit on chain length because, right now, going beyond that is *stupid.*

Cracking time scales with chain length squared. To get an idea of how long the "precalculation" stage of cracking will take, you can ballpark it as follows:

(chain length) * (chain length) * (0.5) / (GPU speed on the hash in question)

Assuming 400M for a GPU (which a GT200 series can do - Fermis are faster, but I don't want to require a Fermi to use this):

Chain length 100,000: 12.5s of regen
Chain length 200,000: 50s of regen
Chain length 1,000,000: 1250s of regen (~21 mins)
Chain length 1,000,000,000: 1,250,000,000s of regen (39.5 years)

So, that chain length is stupid.

[kevinkk5]

yes you´re right.. but the maximum chainlength right now was somehow to low to get good cracking chances with the RT i think.

then i would need 100,000,000 Chain count with 1,000,000 Chain len to get a success probability of 99,57% with md5 len 7 mixalpha-numeric and it would be 1,49GB big...

with len 8 i´d need 10,000,000,000 Chain count and 149GB to get 99,82%!
that´s a bit of space but acceptable

[Bitweasil]

yes you´re right.. but the maximum chainlength right now was somehow to low to get good cracking chances with the RT i think.

Please support your claim.

Code: Select all

#define CHAIN_LENGTH_WARN_MIN 10000
#define CHAIN_LENGTH_WARN_MAX 1000000


These should be warnings - not hard errors. It will allow you to use absurd lengths, but they're stupid to use.

What chain length do you think would be sane, that my code does not currently support? Please back this with math on the cracking/chain regeneration side. A rainbow table that takes 6h per hash to use is not of much use.

[Bitweasil]

yes you´re right.. but the maximum chainlength right now was somehow to low to get good cracking chances with the RT i think.

then i would need 100,000,000 Chain count with 1,000,000 Chain len to get a success probability of 99,57% with md5 len 7 mixalpha-numeric and it would be 1,49GB big...

with len 8 i´d need 10,000,000,000 Chain count and 149GB to get 99,82%!
that´s a bit of space but acceptable

149GB is not an issue anymore... you do realize that you can buy 2TB hard drives cheap, yes?

[kevinkk5]

i got a hard error..

according to winrtgen the maximum cryptanalysis time with a chainlen of 1,000,000 and a Chain Count of 40,000,000 is 4 days (with Cpu).
The success prbability is 97,76%
of course it won´t get better if the chain length get´s higer.

i did not really think too much about the cryptanalysis time. so you are right (like always )

to your quote: that was because i took to low chain count.. that was not your fault

yes i realized that you can buy a 2TB hdd cheap so i said that´s acceptable but it´s a problem if you want to share many tables over the internet.
who downloads tables with more than 100GB... that takes weeks

[Bitweasil]

yes i realized that you can buy a 2TB hdd cheap so i said that´s acceptable but it´s a problem if you want to share many tables over the internet.
who downloads tables with more than 100GB... that takes weeks

*shrug* I don't intend my tables to be downloaded over the internet once they get past 50GB or so. I intend them to be shipped around via mail services on hard drives. Sneakernets are still very valid ways to deal with large amounts of data.

Full US charset len 8 tables will be large. This is unavoidable. So just ship drives. I see no problems here.