[Bitcoin-ml] Time to get agreement on new DAA for Bitcoin Cash

[G. Andrew Stone]

I think the difference is that your DAAs are looking at the the error and
then basically computing the difficulty that would have fixed that error
given the poisson distribution.  I hadn't considered doing it that way and
in that situation I agree with you.  You've introduced the poisson
distribution into your DAA calculation, and so you need to change that to
reflect the new distribution.

The DAAs I thought about are using the error to apply an adjustment to the
prior difficulty.  So the poisson distribution is not needed in the DAA.

On Tue, Oct 31, 2017 at 5:08 PM, Scott Roberts <wordsgalore at gmail.com>
wrote:

> Long tail: If I use effective_D in the simple rolling averaging or WT
> with N=30, my solve time is 8% too low, apparently for the reasons I
> predicted.  If I use the unadjusted D I get solvetimes 8% too high. So
> I wonder what your DAA is. WT-144 gave worse results, 15% error during
> if there are frequent hash attacks.  If I use the unadjusted D and
> multiple by 0.92 for the simple algo or WT with N=30, solvetime is OK.
> I used a simpler approximation that should give better results, but a
> few more long tails.  I just said "if ST > 2x, then D=1/2".   Long
> tails > 4x were 6x fewer, going from 2.4% to 0.38%. So it seems to be
> a great idea. I ran it through attack scenarios and did not see any
> problem.
>
> Simple rolling averages with 1% the hash power of their big brother
> clones have < 1% error in solvetime, so no adjustment routine for
> solvetime is needed unless an asymmetry was thrown in for some reason.
> It's possible for hashrate changes (non-poisson effect) on the order
> of < 2N to cause 20%  longer solvetimes like you say, but due to miner
> profit motive, I don't see that kind of error unless the algo is
> wrong. Saying the simple avgs require poisson is not exactly right:
> they only require a linear relationship between D, ST, and HR.  Since
> your long tail is not symmetrical, it's not linear, hence the ST error
> in the DAAs I have.
>
> I'm interested in your solvetime adjustment, but I can't follow your
> psuedocode.  The only method I can think of is to use a second
> difficulty window of say 10 N  to more slowly modify the fast window
> to keep solvetime on track.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linuxfoundation.org/pipermail/bitcoin-ml/attachments/20171101/9b43c60f/attachment.html>