Le dimanche 19 octobre 2014 à 11:37 -0700, Patrick O'Leary a écrit : > On Sunday, October 19, 2014 12:55:08 PM UTC-5, Milan Bouchet-Valat > wrote: > Le dimanche 19 octobre 2014 à 13:14 -0400, Stefan Karpinski a > écrit : > > That might be why Python has a special function for reading > from it – > > specialized caching behavior. > Note that for those on bleeding-edge Linux distributions, a > new system > call has just been added in 3.17 to get random numbers: > http://lwn.net/Articles/606141/ > > But my understanding is that for scientific applications (as > opposed to > cryptographic ones), /dev/urandom isn't the best choice since > reading > from it is quite expensive for the system due to security > requirements, > and it costs entropy for programs which may really need it. > > It uses a CSPRNG rather than non-CS PRNG, so it is more expensive, and > unnecessarily so if you're not doing crypto. But I don't think the > latter assertion is true--as long as enough entropy got into the pool > in the first place, and no one is trying to use /dev/random (which > they shouldn't be; see for instance > http://www.2uo.de/myths-about-urandom/). I guess you're right: that will only affect people using /dev/random. But (even if that's useless), it's apparently the case of some programs, like GnuPG when generating a key (admittedly not a very common operation).
Regards
