Re: some comments/suggestions on draft-gont-6man-non-stable-iids-02

[Dave Plonka <[e-mail]>]

I totally agree.

On Mar 21, 2018 12:13 AM, "Brian E Carpenter" <[e-mail]>
wrote:

> below...
> On 20/03/2018 11:58, Dave Plonka wrote:
> > Hi Brian and Fernando,
> >
> > On Mon, Mar 19, 2018 at 10:27 PM, Brian E Carpenter
> > <[e-mail]> wrote:
> >> On 20/03/2018 09:17, Fernando Gont wrote:
> >>> Hello, Dave,
> >>>
> >>> Thanks for the feedback!
> >>>
> >>> On 03/19/2018 03:48 PM, Dave Plonka wrote:
> >> ...
> >>>> This is why I propose words along these lines:
> >>>>
> >>>>    A Temporary Interface Identifier MUST be unpredictable by an
> >>>> outside entity even having knowledge of the host's method of
> >>>> generating the Interface Identifier, e.g., pseudo-random number
> >>>> generation algorithm, or knowledge of the host's prior or other
> >>>> Interface Identifiers.
> >>
> >> I know theoreticians who would pick holes even in that formulation.
> >> If I know the weaknesses in your PRNG, and how it picks its seed,
> >> I might be able to predict its output with a finite probability of
> >> success. But that seems like an acceptable risk.
> >
> > I agree. I took the "outside entity" to mean someone that didn't know
> > the seed, but might know the algorithm that operates on the seed.
> >
> > I mentioned PRNG as only one example ("e.g.") method of what an
> > implementer might do.
> > I'd actually prefer the draft not mention randomness, but rather that
> > an IID be unpredictable.
> >
> >>> This would be in replacement of this:
> >>>
> >>>    5.  It must be difficult for an outside entity to predict the
> >>>        Interface Identifiers that will be generated by the algorithm,
> >>>        even with knowledge of the Interface Identifiers generated for
> >>>        configuring other addresses.
> >>
> >> It seems to cover the same ground.
> >
> > They are close, but we could strengthen the implementer's goal from
> > "difficult [to predict]" to "unpredictable."
> >
> > This is inline with Brian's comment that a poor PRNG's output may be
> > still remain difficult to predict, yet have a high probability of
> > success.
> >
> > I certainly don't disagree with what the draft is trying to say, I
> > just think it can be stated better by:
> > (a) not mentioning IIDs differing from prior values or (b) that they
> > be random. In (a), requiring it to differ compromises the randomness
> > because it makes them, in small part, a function of the prior value.
> > In (b), there is no way to know that an IID is random, only plausibly
> > so (by testing it as a member of a set as we say here:
> > https://arxiv.org/pdf/1707.03900.pdf Section 2.1 "Identifying
> > Plausible Randomness," for instance.
> >
> > Perhaps a goal of being "unpredictable" and "plausibly random" would
> > be a good solution.
> >
> > What do you think?
>
> If you mean me personally, saying should or must be unpredicatble is
> necessary and sufficient. If we use the words 'random' or 'pseudorandom'
> you tend to wake up the mathematicians to no purpose.
>
>    Brian
>