On 10/10/2013 03:41 AM, Paul Mackerras wrote: > On Wed, Oct 09, 2013 at 08:07:35AM -0700, H. Peter Anvin wrote: > >> consider the PowerPC random number generator[1]) and > > [snip] > >> [1] which has a known first-order bias which they "correct" for by >> XORing two datums together in a very simple data reduction step. > > 65 actually, not two. > >> However, if their random source has bias it is extremely likely it also >> has nonzero correlations, which require stronger reductions. It would > > The correlations are essentially zero, by design, and experiment > confirms it. Did you see my mail on the kvm list where I explained > how it works? > No, sorry... I got a bit of detached discussion as part of benh talking about KVM and randomness (for the record, I'm all for better randomness on all platforms.) Either way, XORing samples is a pretty inefficient (both in terms of anticorrelation and in terms of entropy efficiency) form of data reduction/conditioning. It would still be better to feed the output into the pool with a 65x derating. -hpa -- To unsubscribe from this list: send the line "unsubscribe linux-doc" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
