On Wed, 13 Mar 2024 at 14:34, Kent Overstreet <kent.overstr...@linux.dev> wrote:
>
> I liked your MAD suggestion, but the catch was that we need an
> exponentially weighted version,
The code for the weighted version literally doesn't change.
The variance value is different, but the difference between MAD and
standard deviation is basically just a constant factor (which will be
different for different distributions, but so what? Any _particular_
case will have a particular distribution).
So why would a constant factor make _any_ difference for any
exponential weighting?
Anyway, feel free to keep your code in bcachefs.
And maybe xfs even wants to copy that code. I don't care, it seems
stupid, but that's a filesystem choice.
But if we're making it a generic kernel library, it needs to be sane.
Not making people do 64-bit square roots and 128-bit divides just for
a random statistical element.
Linus
