[onerng talk] RNG designs

[Jim Cheetham]

On Sun, Feb 2, 2014 at 8:27 PM, Paul Campbell <paul at taniwha.com> wrote:
> yes I'd prefer that the default behaviour did something useful (ie it's a
> stream the drng will consume without complaint) even if it's not 'perfect' in
> some form - personally I think the RF noise source is somewhat more "suspect",
> I don't think it should be part of the default

But even if the RF stream delivers less than 1 bit of entropy per bit,
that doesn't really hurt us. It is still better than zero :-)

If the diode circuit delivers a better result, is it really 'perfect'?

>From reading of the Turbid documentation
(http://www.av8n.com/turbid/paper/turbid.htm#sec-hrng starts half-way
through) the hashing stage is what mitigates problems in the input. I
don't see that mixing all the sources into a single pool
(interleaving, xor-ing; does it matter which?) and then hashing them
before presenting them to the user can be a problem.

>> This is all software/firmware we're talking about, isn't it?
>
> firmware (we can't really change this for most users after we ship)  - so far
> I'm not planning on writing any host side software (other than maybe a udev
> script)

Our default firmware should be the most useful, but not necessarily
the most flexible. People who want flexibility should be able to get
the programmer and play with alternative firmwares, or preferably
write their own.

-jim