On Fri, 17 Aug 2007, Winfried Ritsch wrote:
But since one rule is, never guess, measure why, I have a quick question. Is there a profiling tool on oject or message base ? I can remember there was one in Max/FTS, so I can find out the bad objects... ?
Since I reimplemented the stack overflow checker so that it really checks every sent message, it's now easier to make a profiler. Actually, I just made one tonight, when reading your email. It looks like this:
http://artengine.ca/desiredata/gallery/simple-benchmark.png
those green numbers represent real time, measured hierarchically (that is, all sent messages count as part of the sender), whereas it could have excluded sent messages so that all visible times add up (either is more useful than the other sometimes). It is an exhaustive measurement, that is, every message is counted, whereas it could be statistical instead, using unix alarms to make statistics on stack snapshots.
You make the green numbers appear by selecting "get elapsed" from the View menu.
This example was made from making a few clicks on the messagebox and it is measured such that 1 unit = 800000000 cpu ticks. in practice my cpu switched to a higher speed. Correct profiling involves setting the speed of the cpu to a fixed value so that the values can be displayed in actual seconds.
If you want to try it out we could meet online and set it up on your computer.
_ _ __ ___ _____ ________ _____________ _____________________ ... | Mathieu Bouchard - tél:+1.514.383.3801, Montréal QC Canada