Re: Review: Revise parallel pg_restore's scheduling heuristic

On Tue, Aug 04, 2009 at 10:45:52AM -0400, Tom Lane wrote:
> Sam Mason <sam(at)samason(dot)me(dot)uk> writes:
> > t = 0.54 ((avg1 - avg2) / (stddev * sqrt(2/samples)))
>
> > We then have to choose how certain we want to be that they're actually
> > different, 90% is a reasonably easy level to hit (i.e. one part in ten,
> > with 95% being more commonly quoted). For 20 samples we have 19 degrees
> > of freedom--giving us a cut-off[1] of 1.328. 0.54 is obviously well
> > below this allowing us to say that there's no "statistical significance"
> > between the two samples at a 90% level.
>
> Hmm, so what about 95% or 99% confidence?

The cut-off goes up to 1.729 for 95% and to 2.539 for 99%. These values
are only really for a 20 samples with the above calculation, the link I
gave above gives a nice table for different values.

I've also realized that I did the standard deviation all wrong. I
should have calculated them independently and then got the mean:

stddev1 = 159.9699
stddev2 = 129.6466
stddev = 144.8083 ((stddev1+stddev2) / 2)

Here it makes absolutely no difference, but when they were really
different distributions it would.

--
Sam http://samason.me.uk/