Re: update on the djbdns bugs? (fwd)

On Wed, Mar 10, 2010 at 3:40 PM, Dean Anderson <[email protected]> wrote:
>> Dean, I was using simple roulette logic.  65,536 ports * 65,536 qids /
>> 200 simultaneously valid responses = 21M.  We're dealing in orders of
>> magnitudes, it's a reasonable approximation.
>
> You still seem to be channeling Palin:
>
> (%i69) 65536 * 65536 * 200;
> (%o69)                           858993459200
>
> Quite a bit higher than 21M.

For anyone thinking Dean deserves any credibility on this issue, take
a look at just this simple example: Dan Kaminsky is pointing out that
65536 * 65536 / 200 = 21M, but Dean counters by pointing out that
65536 * 65536 _*_ 200 != 21M.

> But you have no credible explanation for your math.

Further, the math behind Dan's number is straight forward: an attacker
has to guess two 16-bit numbers (i.e., the 65536 * 65536), but there
are 200 winning pairs to choose from (i.e., the 200 outstanding
queries for identical names).  Having more winning pairs means the
number of expected guesses before success *decreases* (i.e.,
division), not increases (i.e., multiplication).