On 5 May 2005, at 09:18, Melchior FRANZ wrote:
I don't have much time to delve into this right now (hah!), but I have some log calls in my local build, which have been fairly consistent (haven't tested since your changes, Melchior - when I do, I'll post those numbers).
That said, here are the numbers I have (for three runs)
time to run fgMainInit:
time between idle_state 0 and idle_state 1000
(wall-clock) time after hitting idle_state 1000
before scene appears
consistently about 25 - 30 seconds
Comments - the first phase (lots of database traversal) is obviously very dependant on file system cache hot-ness, and the second phase similarly. The third phase, *after* idle-state 1000, is the bit I was referring to when I talked about starvation; while it's doing this wait, I see the splash screen, and see log output from subsystems frequently (traffic manager, clouds, ephemeris), but it seems to sit there for ages before showing the cockpit + scenery.
I ran a statistical profiler on the startup, and it didn't get all the way through startup before it hit it's log size limit, but the first part was spending huge amounts of time parsing data files, especially doing string - to float parsing. Doing the equivalent of 'atof' was something like 11% of the total time for the profile I ran.
Lazyness is the way to go here, I'd suggest. Or binary file formats.
Ignorance more frequently begets confidence than does knowledge.
_______________________________________________ Flightgear-devel mailing list Flightgearfirstname.lastname@example.org http://mail.flightgear.org/mailman/listinfo/flightgear-devel 2f585eeea02e2c79d7b1d8c4963bae2d