[Octave-bug-tracker] [bug #58800] BIST for rng sometimes fails

[Hg200]

Follow-up Comment #20, bug #58800 (project octave):

It is difficult to find literature about correctly MT-seeding and then we also
have the Octave specific implementation.

My nightmare is that if we setup the generator with too few non-zero entropy
numbers, that the generator will take several cycles until it produces
reliable random patterns. I use Octave for Monte Carlo simulations and I don't
want any degradation in the random number quality. On closer investigation of
randmtzig.cc, it looks as if the MT-state is set up with the seed 19650218UL.
Entropy is added later. So my understanding is that we still should get
reliable random numbers even if we don't add entropy at all. Therefore, there
should be no loss of statistical number quality if we remove "urandom" and
use, for example, only the "system clock" instead.

The question remains whether the order or the position at which the entropy is
given matters. In the patch, clock and random numbers are exchanged compared
to earlier. I don't believe it has any impact, but i don't know fur sure.

Finally I agree to Markus' argument that the number of possible
"entropy-realizations" determines the number of different "random-number
realizatins". However: ".. if the seeds are different, the generated
pseudo-random number streams will have no detectable relation ..". Here i
would ask the question if this also holds for the hamming distance of the
output if two different seeds come close together (but i don't have an answer
for this as well).

    _______________________________________________________

Reply to this item at:

  <https://savannah.gnu.org/bugs/?58800>

_______________________________________________
  Message sent via Savannah
  https://savannah.gnu.org/