[EMAIL PROTECTED] (Ton Hospel) writes:
> I also believe it still takes averages of speeds, but a good bench should
> take averages of times (and convert them back to speeds for presentation
> possibly). This is standard, and makes sense, since if you have a program
> that uses all things tested, it's average time that counts, while average
> speed has no real meaning.
I fail to see how the average can have any real meaning anyway. It is
just an average of random snippets of code which basically tries to
test one feature in one way each, and for enough time to make it
Whatever you do the average will not be able to predict how fast a
normal application would run. We might as well remove it.
But it is kind of nice to end up with a single number that tells the
whole story :-)