Re: random .. ?

From: Cristiano (cristiano.pi_at_NSquipo.it)
Date: 02/02/05 

Date: Tue, 01 Feb 2005 23:06:52 GMT

David Wagner wrote:
> Cristiano wrote:
>> I'd say that DH seems better-suited for testing 32-bit pseudo-random
>> generators (once you fix all the bugs).
>
> IMHO, DIEHARD is not very useful for testing 32-bit PRGs. It can find
> grotesquely broken PRGs, but there are other better ways to find
> grotesquely broken PRGs. It is a fine sanity check just to make sure
> you haven't overlooked something obvious (after all, it is easy
> enough to run DIEHARD on your PRG output), but passing DIEHARD is
> much too low a bar. We can do much better.

Agreed. One uses a statistical test because it is easy and to avoid a deep
analysis of the generator.

>> For bits and physical generators it should be preferable to use a
>> tester which needs a reasonable amount of bits (obviously I suggest
>> RaBiGeTe).
>
> Ok. I believe you. I don't have experience with RaBiGeTe vs DIEHARD.

I have no doubt; I think only 2 or 3 people know RaBiGeTe. :-)

Cristiano

Relevant Pages

  • Re: Interesting
    ... As I don't have the bits around to count radioactive decay (which I believe to be about as good as it gets for RNG) I'll have to look at mathematical trickery. ... I also previously posted a link to the source for Dr Marsaglia's "KISS" generator which has a period of something like 2^123 and does well on his battery of tests. ... Two 16-bit multiply-with-carry generators, ... As for the comment on radioactive decay bits and mathematical trickery, I'll simply quote a short section from Dr Marsaglia's intro to the contents of his Diehard CD... ...
    (microsoft.public.vb.general.discussion)
  • Re: AES and Diehard
    ... > Cristiano wrote: ... >>You should post less and read more (this is just my suggestion). ... >>know if this is good English, but I hope you understand its meaning. ... IV in the output that was analyzed by Diehard. ...
    (sci.crypt)
  • Re: best approach to generate random number in java
    ... Pseudorandom number generators are ... > Some PRGs give results with better distributions than others. ... Therefore saying "more random" is like saying "more perfect". ...
    (comp.lang.java.programmer)