[Cryptech Core] deeper entropy analysis

[Leif Johansson]

On 12/10/2014 08:07 PM, Randy Bush wrote:
>> Some time ago I was wildly confused by my DMA based entropy sources would 
>> prefer alternating bit patterns over repeating ones, as I'm sure you've seen 
>> on the tech list.
>>
>> I asked someone with considerably more knowledgeable than me about how one 
>> could perhaps analyse the output to gain insights into how this could be. That 
>> person did not have time to contribute to the problem, but he suggested we 
>> could perhaps involve this guy (for paid work)
>>
>>   http://www.csc.kth.se/~terelius/
>>
>> I haven't talked with him at all, but while I think we can be confident that 
>> our TRNG will output excellent random data, if we want to be as confident about 
>> the entropy I think we will need help from someone with higher mathematical 
>> skills than I possess  - possibly that guy.
>>
>> If you think it is a good idea then maybe I can try to meet with him for lunch 
>> or something and see if he is interested and see what he could do for
>> us?
> 
> as you say, whiten it, mix it, teach it to chacha, and we'll probably
> have good bits.  but better understanding of the input would be good.
> and it might reduce the chance of late surprises, one hopes.

right

my thinking is that since we've added all those measurement points in
the chain, folks will eventually measure and then it would be great to
_at least_ have some idea what characterizes really bad noise on the
input side of the dance instructor or folks will panic over every
statistical anomaly they see

> 
> it's research, so let's not poke too hard at the upsides.  though it
> would be good to make the research goal(s) clear with terelius.
> 

fully agree

> are there downsides?  your time?  presumably funding if lief can not get
> terelius's time donated?
> 

Let me try to poke at KTH and see what they say and we can decide then.

	Cheers Leif