Re: [gentoo-user] Re: Re: Re: OT: Mapping random numbers (PRNG)

[Matti Nykyri]

On Sun, Jun 29, 2014 at 02:38:51PM +0200, Kai Krakow wrote:
> Matti Nykyri <matti.nykyri@×××.fi> schrieb:
>
> > That is why the possibility for 0 and 1 (after modulo 62) is twice as
> > large compared to all other values (2-61).
>
> Ah, now I get it.
>
> > By definition random means that the probability for every value should be
> > the same. So if you have 62 options and even distribution of probability
> > the probability for each of them is 1/62.
>
> Still, the increased probability for single elements should hit different
> elements each time. So for large sets it will distribute - however, I now
> get why it's not completely random by definition.

Usually when you need random data the quality needs to be good! Key, 
passwords etc. For example if an attacker knows that your random number 
generator same or the next index with double probability, he will most 
likely crack each character with half the tries. So for each character 
in your password the time is split in half. Again 8 character password 
becomes 2^8 times easier to break compared to truely random data. This 
is just an example though.

> > Try counting how of often new_index = index and new_index = (index + 1) %
> > 62 and new_index = (index + 2) % 62. With your algorithm the last one
> > should be significantly less then the first two in large sample.
>
> I will try that. It looks like a good approach.

Ok. I wrote a little library that takes random data and mathematically 
accurately splits it into wanted data. It is attached to the mail. You 
only need to specify the random source and the maximum number you wish 
to see in your set. So with 5 you get everything from 0 to 5 (in total 
of 6 elements). The library takes care of buffering. And most 
importantly keeps probabilities equal :)

-- 
-Matti