[ale] OT - making really strong pass phrases - was New encryption technology using a piece of paper

[Tim Watts]

Interesting stuff. Unless I'm misunderstanding this, these times only
apply under the assumed conditions (multiple 1-4 char words etc.). If
the off-line attackers only have say a 256-bit hash of your
password/phrase then they have no rational basis to make these
assumptions. Depending on the attacker's degree of motivation and
sophistication they might just try pure brute force. If they have vast
resources I suppose they could run multiple server farms each applying a
different crack strategy in parallel. More than likely I would think,
they're going to allocate a only certain amount of time for each
password (or the database as a whole) applying ever more sophisticated
crack strategies ranging from simple human patterns to pure brute force.

None of this, of course, is arguing against preparing for the worst
case. But I wonder: if they have limited resources (time, hardware) how
could the attackers decide that a quasi-brute force strategy (e.g.
assume 1-6 OPIE words) is more likely to be effective than say a pure
brute force one? My sense is that they can't and so their choice is kind
of another random element in the attack.

Also, let's not forget that those times are upper bounds. There's always
the chance they could get it on the 1st try or anywhere in between!


On Tue, 2011-09-06 at 20:27 -0400, Ron Frazier wrote:
> Michael W.,
> 
> OK.  I'm impressed.  Assuming I did the math right, crack time is 98 
> thousand years with a 1000 pc botnet.
> 
> You've pretty much convinced me to use long simple pass phrases if I 
> have a choice, unless the website or application won't accept it.
> 
> Thanks for the info.  Thanks also to Michael T. and others who joined in 
> the discussion.
> 
> Here are some numbers I thought everyone might like to consider.  
> Estimated offline crack time based on 1000 pc botnet running at 100 
> trillion guesses per second.  Using a 2048 word lexicon and simple pass 
> phrase, giving 11 bits of entropy / word.  Bigger answers are all in 
> days.  To get these numbers (in days), take the power of 2 (# of 
> permutations) and divide by 8.64 x 10^18.
> 
> 2 words - 2^22 permutations - 42 NANOSECONDS
> 3 words - 2^33 permutations - 86 MICROSECONDS
> 4 words - 2^44 permutations - 176 MILLISECONDS
> 5 words - 2^55 permutations - 360 SECONDS
> 6 words - 2^66 permutations - 8.54 days
> 7 words - 2^77 permutations - 17.49 thousand days = 47.92 years
> 8 words - 2^88 permutations - 35.82 million days = 98.14 thousand years
> 9 words - 2^99 permutations - 73.36 billion days = 200.98 million years
> 
> My take away from this is: if you want protection from a botnet, don't 
> even consider a pass phrase less than 6 words if using a 2048 word 
> lexicon.  If you only want protection from a fast attack by a single 
> machine or small GPU array, multiply these crack times by 1000.  Pass 
> phrases 5 words and less for this purpose are almost worthless.
> 
> Sincerely,
> 
> Ron
> 
> 
> On 9/6/2011 5:17 PM, Michael H. Warfield wrote:
> > Ah...  That's the whole point.  Yes you can go down this road and add
> > complexity (and misery) to the process but you can accomplish the same
> > task by adding words that are easy to read and process and much easier
> > to support.
> >
> > Do the math again for 8 words.  88 bits of entropy.
> >
> >    
>