As I have pointed out already, I never claimed a normal distribution.
I never claimed any kind of distribution that follows any mathematical
model at all. All I claimed was that I had looked at a large number of
results from a bunch of different tests, plotted them, and discovered
that the min wasn't a very useful number. In fact, I found that it was
so useless that, in my opinion, it was unethical to report the min.
: That argument seems correct under the assumption that the variable
: is normally distributed .. i.e. I think you'd get a wildly varying
: absolute result from a mimimum of a number of tests from a normal
: curve. Not that I thought about that for more than five secs ..
:
: If the variable is not normally distributed, but from something like
: x^n * exp(-nx) instead, then the minimum would tend to zero with certainty.
: I.e. you'd be sure to really get the minimum in the end. In fact,
: that's obviously true in general, no matter what the distribution. So
: again, looks like Larry is wrong there. Coincidence twice!
I'm sorry, are you saying that benchmarks results all follow some model
where the min is the right answer? Could you explain that please?
What if the distribution is normal? What if 95% of the time you get
one value and 5% of the time you get a smaller one, are you saying that
the smaller one is the right answer?
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/