hardware accelerated random()

[Yann Droneaud]

Le dimanche 09 décembre 2012 à 21:41 +0100, Ondřej Bílka a écrit :
> Hello, 
> I wrote random generator that uses aes-ni. I generate random numbers in
> groups of 16 and then use them.
> On my i5 test program spend about 6.4 cycles per call. 
> I put sample implementation on
> http://kam.mff.cuni.cz/~ondra/rand_aes.tar.bz2
> 
> I am not cryptographer so I do not know if this ways causes significant
> statistical weakness.
> 

Without going to the crypto strength PRNG, which are not required for
libc, do you have so demonstration of the quality of this random
generator ?

Check those pages for some tools to validate a PRNG:

https://bitbucket.org/haypo/hasard/wiki/Home#!test-rng-quality
https://gitorious.org/randbuf/randbuf/blobs/master/VALIDATE

http://www.random.org/analysis/


PS: you may find some background information in the following article
"LCE: Don't play dice with random numbers" by Michael Kerrisk, based on
a talk by H. Peter Anvin https://lwn.net/Articles/525459/

Regards.

-- 
Yann Droneaud
OPTEYA