Security Dynamics

[Eric Hughes]

Re: checking distribution in 10^4 samples

>their simulations were based on five to ten runs, with 10^7 updates
>per run.  they aren't precise about the exact number of random
>numbers needed, at least not in this paper, but i assume it's in the
>order of one per update, in which case 10,000 would not be enough.

The method of randomness-checking done here is to run a physical
simulation with the random numbers.  Direct statistical methods are
much more efficient.

Eric