[plt-scheme] V301.5 Speed Up

On 2/10/06, Noel Welsh <noelwelsh at yahoo.com> wrote:
> This is a philosophical issue.  Do we want to measure the
> best we can expect when the stars align, or what the user
> can expect on a normal day?  I choose the latter.  You
> could equally choose the former.  It is up to personal
> choice.

I want to measure my program, not my program as it interacts with the
umpteen million other things running on my machine.  That's why I'm
interested in the former measure.  I don't know what a `normal day'
is, but I *would* like to know the minimum amount of time I'd have to
wait.

> On Gaussian noise:  I would expect, for the reasons Greg
> gives, the distribution to be gaussian.  One could test for
> this.  I haven't, because I haven't implemented the code to
> do so.  However it would be simple to confirm this by eye
> by running a 1000 benchmarks and eye balling the data
> (using the histogram plotting functions in the science
> collection of course).

I bet it is much more complicated than a gaussian.  I can think of
several factors off the top of my head that could make it non-gaussian
(a process switch at an inopportune moment could make the distribution
severely bimodal).  I think a demonstration that it *is* gaussian is
called for.


> If the running time was very small then this assumption
> might be in danger, as the gaussian is symmetric but
> obviously running time can't be less than zero.  In this
> case a Poisson might be appropriate.

I'd imagine the distribution to be scale-free in the running time
unless the running time were on the same order of magnitude as the
clock cycle.

> As for discontinuities in the data, I'm not sure how much
> of a worry they are.  It seems the timer has a resolution
> of 10ms on my system, but if the running times are large
> enough  relative to this resolution it shouldn't matter.
> Again, this assumption should be tested.

Yes.  I'm curious.

--
~jrm