<p dir="ltr">Also, for such a short test, preemption is also likely a factor. Meaning another task might get scheduled in between and thus distorting your perception of time.<br>
For proper performance testing you either need to use some other metric rather than time, or you need to have your test run long enough so that variance like this would be insignificant..</p>
<p dir="ltr">/Jesper</p>
<div class="gmail_quote">Den 3 apr 2015 5:35 em skrev Wolfgang Rostek <wolfgang.rostek@gmx.de>:<br type='attribution'><blockquote class="quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div>
<div>...<br />
> > Frequence scaling I can't imagine on my dektop machine.<br />
> > It is a jump between 350ns and 800ns. In the loop below<br />
> > all the events of one test run are of equal time distance,<br />
> > either slow or fast.<br />
> <br />
> Frequency scaling can still definitely play a role, even on a desktop machine.<br />
> Also, Core i5 chips support Turbo Boost which should be disabled to<br />
> perform benchmarks.<br />
> <br />
> I'm guessing you get your time measurements from the timestamps in the<br />
> trace?<br />
Yes<br />
<br />
I'm not hunting for a detailed benchmark. Only such a large variance makes<br />
me thoughtful. For such a test I did expect some 10-30% but not this more <br />
than 200%.<br />
<br />
Wolfgang R.<br />
<br />
_______________________________________________<br />
lttng-dev mailing list<br />
lttng-dev@lists.lttng.org<br />
<a href="http://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev">http://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev</a><br />
</div>
</div>
</blockquote></div>