> -----Original Message----- > From: Dave Sherohman [mailto:[EMAIL PROTECTED] > Sent: Thursday, October 25, 2001 12:58 PM > To: debian-user@lists.debian.org > Subject: Re: More accurate time feedback > > > On Thu, Oct 25, 2001 at 09:06:29AM -0700, Andrew Agno wrote: > > Dave Sherohman writes: > > > faster than a quicksort.) If you are required to use a > small data > > > set, just perform more repetitions to get measurable data. > > > > This also allows you to obtain statistics like standard > deviations and > > confidence intervals. > > Umm... No. If you run 10,000 reps and only measure the total time > (because each rep is too quick to be measured individually), you can > only determine the mean time per rep. You'd need the individual rep > times to derive any other stats, unless there are some fancy > techniques that I'm not aware of. (If you do 2 reps and only measure > that the total time is 10 ms, you have no way of knowing whether the > individual times were 5 ms and 5 ms or 1 ms and 9 ms. This makes > computing anything other than the mean rather difficult.) >
You might run many batches of 1,000 reps. The statistics of that population would be useful, if the error in the time measurement is small enough.
