Re: [radext] WGLC #2 for draft-ietf-radext-dtls-04

[Alan DeKok <[e-mail]>]

Sam Hartman wrote:
> I don't know.  Enumerating through I think 96^11 md4s is apparently
> trivial given reasonably-priced hardware for cracking.  So probably $10k
> investment.  See recent discussions of hash cracking on GPUs.  However,
> md5 is a bit slower, and 11 is less than 16.

  http://www.insidepro.com/eng/egb.shtml

  Says that they can do about 500M hashes/s.  Let's assume that's true,
and is the best case.  This is even though other sites claim 100x that
cracking power.

  These numbers are for *password* cracking, which is a bit different
from cracking RADIUS packets.  So maybe RADIUS will be slower.

>  I think I'd say that I
> would be unsurprised if someone broke RADIUS secrets entirely, and would
> be entirely unsurprised if brute forcing the secrets anyone is actually
> using in practice was easy.

  I agree.  Let's say the MD5 cracking of RADIUS packets was 1% of the
above password cracking speed.  Even with that, brute-forcing all
alphanumerical passwords would take less than a year for one person.

  I have 94 individual characters available on my keyboard.  Let's say I
use them all for shared secrets, and that the cracking speed is 5M
attempts per second.  We have the following crack times, for each length
of shared secret:

4	16 seconds
5 	25 minutes
6	38 hours
7	150 days
8	37 years

  If you assume that the attacker can do 500M attempts per second, then
all shared secrets less than 8 characters long can be trivially broken.
 And it gets MUCH worse than that if people are less rigorous about
using a wide range of ASCII characters.

  Alan DeKok.