Re: failed statistical RNG test

[Matthias Drochner <M.Drochner%fz-juelich.de@localhost>]

On Tue, 2 Dec 2014 21:52:35 +0000
Christos Zoulas <christos%astron.com@localhost> wrote:
> In article
> <CABfrOT-QTAT-Hvap5zK2DQ6Vwp1k6zhJAkgTSWBTBh1FyhhfuA%mail.gmail.com@localhost>,
> bch  <brad.harder%gmail.com@localhost> wrote:
> >This just popped up on the console (during
> >./pkgsrc/databases/mysql56-server build):
> >
> >(transcribing by hand):
> >===============================
> >ny runs of 4 0s (386 > 384)
> >cprng 24455 6096 42: failed statistical RNG test
> >===============================
>
> Aha! I know I was not the only one (I turned a DEBUG into DIAGNOSTIC)
> so we can find out what's wrong...

The test target was only missed by a little bit (386 vs 384), so I'd
assume that it was only due to the statistical nature of the tests
and no serious RNG failure.
I had seen such error messages last year; here is what I wrote
about that:
> Yes, and I've come to the conclusion that the errors are within the
> limits to be expected by statistics.
>
> I've done the math for the "monobit" test (because it is just a
> binomial distribution) and found that the probability to have less
> than 9725 ones in 20000 bits is ~4.9e-5. This means that we can
> expect an error rate (<9725 or >10275 ones) of about one in 10000.
> Now I've added a counter for test invocations to the kernel.
> Got 13 hits of the monobit test within 143491 calls.
> (These 140k calls took a minute or two running my test program.)

One could assume that the various RNG tests are designed to have a
similar probability of expected failures. It would be good to know
how often the RNG tests were invoked.

best regards
Matthias


------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------
Forschungszentrum Juelich GmbH
52425 Juelich
Sitz der Gesellschaft: Juelich
Eingetragen im Handelsregister des Amtsgerichts Dueren Nr. HR B 3498
Vorsitzender des Aufsichtsrats: MinDir Dr. Karl Eugen Huthmacher
Geschaeftsfuehrung: Prof. Dr.-Ing. Wolfgang Marquardt (Vorsitzender),
Karsten Beneke (stellv. Vorsitzender), Prof. Dr.-Ing. Harald Bolt,
Prof. Dr. Sebastian M. Schmidt
------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------