On Wed, 24 Apr 2002, Sandy Harris wrote:
> Jim Choate wrote:
>
> > > PRNG output is fixed/repeatable too - that is a properly you *want* from a
> > > PRNG.
> >
> > No it isn't. You -want- a RNG but you can't have one. Nobody -wants- a
> > PRNG, they -settle- for it.
>
> That is nearly true for crypto applications, but it certainly isn't for
> some others. e.g. If you're debugging simulation software, you may need
> to be able to make the PRNG produce repeatable output by giving it the
> same seed on every run.
Granted, test vectors are a good tool but that's more a means to an end
rahter than an end. At least in my mind.
What is the expectation in your example (for example) once the system goes
into use? Does the need for repeatability still apply? I'd guess not. So
we're right back to where we started.
Would you rather run your Monte Carlo with a PRNG or a RNG? I'd say it
depends on a fuzzy concept called 'precision'. How close do you need to
be? The closer you get the more it costs, it's usually not linear either.
> Given a well-designed PRNG, though, it is not clear that there's any
> real benefit to using a true RNG instead.
There's always a cost-benefit-effort tradeoff. Of this I think we all
agree.
--
____________________________________________________________________
The law is applied philosophy and a philosphical system is
only as valid as its first principles.
James Patrick Kelly - "Wildlife"
[EMAIL PROTECTED] www.ssz.com
[EMAIL PROTECTED] www.open-forge.org
--------------------------------------------------------------------
