Hi,

i added the garbage collect and the numbers are better ! There are fluctuations but no more "random-like" variations of the measured values.

I computed the standard error which is, for all measures i did, inferior to 0.5% of the mean (then the confidence interval is good).

Thank you both :)

Steven.

Le 2017-06-14 22:08, werner kassens a écrit :

On 06/14/2017 09:12 PM, Steven Costiou wrote:

repetitions timesRepeat:[ results add: [1000000 timesRepeat:[o m]] timeToRun].

Hi Steven,

the influence of the garbage collector is in cases like these often somewhat difficult to forecast. i would eventually change the code to:

repetitions timesRepeat:[ Smalltalk garbageCollect. results add: [1000000 timesRepeat:[o m]] timeToRun].

and if i wanted to see the influence of the garbage collector i would enlarge repetitions and diminish 1000000 in such a way that the product remains constant. of course one would not want to destroy that influence completely as in the real world it always takes its toll.
werner