> Thanks for your input :)
No problem -- I really do enjoy this. However, I do have some actual
WORK that I need to do today :(
> > Finally, there were serious errors in your methodology in your
>
> Serious? I thought "in the Big Picture, it won't matter a gnats eyebrow." :)
And I stand by that ... Such is the fallacy of the benchmark: "hah!
Technique Foo was ten times faster than Technique Bar!" But, if one
takes 5 microseconds and the other takes 50, they're both going to be
blown out of the water by the first page-fault that crosses their
path, or waiting for an I/O operation to complete, or someone moving
the mouse causing the screen to refresh, or a million other things
that will make that 45 microsecond difference pale in comparison.
>
> > original benchmark. It turns out the printing dominated the total
>
> That is why I made both identical except for the difference I'm
> concerned with.
Still poor methodology -- A thought-experiment. I am testing two
different implementations -- now $DEITY knows that one of them will
take 10 milliseconds to execute, and the other an entire second. (I,
however, not being omniscient don't know that -- hence my benchmarking
test).
I put them into a test-crib with a second function that takes
precisely 24-hours to complete. So, one of them takes 86400.01
seconds, and the other takes 86401.00 seconds. Now, assume that the
24-hour extraneous function really might vary randomly plus or minus
five minutes. Suddenly that variance in the spurious function will
totally swamp the variance I am trying to measure between the two
implementations.
In your benchmark, you did precisely that -- by putting a function
that takes a relatively very-very-long and randomly-variable amount of
time (printing) inside the function you are trying to test, you are
basically destroying the exact number you are trying to compute, and
have merely computed the deviation of the print operator.
-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
Lawrence Statton - [EMAIL PROTECTED] s/aba/c/g
Computer software consists of only two components: ones and
zeros, in roughly equal proportions. All that is required is to
sort them into the correct order.
--
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
<http://learn.perl.org/> <http://learn.perl.org/first-response>
