Messages in this thread |  | | From | (Otmar Lendl) | Subject | Re: CONFIG_RANDOM option for 1.99.2 | Date | 23 May 1996 14:20:57 GMT |
| |
Aaron Ucko <UCKO@vax1.rockhurst.edu> wrote: > >Yes, but nonsecurity-related userspace programs might also want good random >numbers. Consider a simulation using the Monte Carlo method--pseudorandom >numbers might produce misleading results.
.. as might random numbers (whatever that is). But at least you know the risk when using PRN.
Using /dev/[u]random is a *bad* idea, since you want reproducable pseudorandom numbers, and *not* random numbers.
Even just using /dev/[u]random for the seed of a conventional PRNG is IMHO a bad idea, too. The seed shouldn't matter at all for the quality of the PRN, when you're using a decent generator, so you gain nothing in return for losing reproducibility.
No, If you want to do stochastic simulation (like quasi-Monte Carlo) on a *scientific* (that is verifyable) basis, there is no way around using reproducible source of PRN. Just be careful which generator you use. Or even better, do the calculation with at least two completely different PRNG, like a LCG (avoid RANDU like the pest !) and an EICG. See http://random.mat.sbg.ac.at/ or my upcoming master thesis for more info on this.
Summary: Use /dev/random when you need non-guessable numbers, but use conventional PRNG for heavy duty stochastic simulation.
And for configuring Linux: I think we might need some Main option in the config to switch between (fast, convenient and insecure) to (secure at all cost). That choice seems to pop up more that once, and not only in the kernel. (/proc, su, echo/discard in inetd.conf, ...)
Sorry for being mostly off-topic.
otmar -- / Otmar Lendl (lendl@cosy.sbg.ac.at) # http://www.cosy.sbg.ac.at/~lendl/ \ \ Killfiles generate SEP fields. Beware: the CE-Norm does not cover them. /
|  |