[Shootout-list] Ray tracer
Bengt Kleberg
bengt.kleberg@ericsson.com
Mon, 13 Jun 2005 10:30:00 +0200
On 2005-06-13 09:36, John Skaller wrote:
...deleted
> are in fact meaningless, as the following shows: this is the very
> same binary on an unloaded AMD64 giving wildly different times
> on each run: from 2.784 to 3.577 in just 4 samples.
the numbers you have gives me 3.2 seconds, +- 12%. while not very good i
still find +-12% to be meaningful. had you reached +-50% i would agree
about meaningless. +- 30% i would have labeled not usefull.
...deleted
> This shows that there is absolutely no such thing as
> an 'unloaded' machine. The differences here are
> entirely due to cache effects from random kernel
> activity.
while i do not know enough about this particular computer i can still
wonder if all the differences are random. perhaps network activity plays
a part? or cron activity? neither is totally random. imho.
i think one conclusion from this is that it is unreasonable to use only
the slowest time for each ''n''. perhaps an average would be better?
what are the alternatives?
bengt