Re: Poll Results

["muppet" <scott asofyet org>]

Torsten Schoenfeld said:
> On Mon, 2005-01-31 at 22:28 -0500, muppet wrote:
>
> > http://asofyet.org/muppet/software/gtk2-perl/startup-time-poll/
>
> | there were not enough data points from multiple versions on a single
> | machine
>
> I think that's where my sandboxes come into play.  I can only offer
> typical runs though:

ooooooooooh, thanks!

reformatting slightly:


   Gtk2       gtk+     # types   load time
   1.072      2.0.6     295      0.043s
   1.072      2.2.1     299      0.048s
   1.072      2.2.4     299      0.048s
   1.072      2.4.7     335      0.056s
   1.072      2.4.14    335      0.057s
   1.072      2.6.2     343      0.054s

I could see a trend right up till the end.  (#types is the number of types
listed in the maps-2.* files.)


   Glib       glib      load time
   1.072      2.0.6     0.014s
   1.072      2.2.1     0.020s
   1.072      2.2.3     0.021s
   1.072      2.4.6     0.019s
   1.072      2.4.8     0.020s
   1.072      2.6.2     0.016s

That one blows it even more.

   Gtk2->init
   0.014s
   0.016s
   0.014s
   0.017s
   0.016s
   0.022s

This looks reasonable.  As it deals with IPC i'm not inclined to draw any
conclusions from it.


    total startup time
    0.071s
    0.084s
    0.083s
    0.092s
    0.093s
    0.091s

Less than a tenth of a second in all cases.  Congratulations!


It looks like we can't really sort out the time to initialize those data
structures from the noise of the system timing.  All sorts of factors play
into timing; system load, page faults, power surges, sunspots, so meaningful
stats on numbers so small are hard to get.  I'm happy that we have this
problem, because it means the startup time is Good Enough.


-- 
muppet <scott at asofyet dot org>