DEC
30
2005

Simplistic KDE performance numbers

Hiya,

here's a *very* simple way to get a performance number for KDE:

$ time konsole -e sh -c exit

This gave me the following numbers:

0.75 seconds on a notebook, Sempron 3000+, SUSE 10.0
0.80 seconds on a desktop, Athlon XP+ 2000, Slackware 9.1, KDE compiled from svn
1.0 seconds on a notebook, Intel P4 M, 1.8 GHz, SUSE 10.0
1.0 seconds on a notebook, Intel P4 M, 1.8 GHz, kUbuntu 5.10

(as a comparison xterm takes 0.1 seconds for this).
I guess we could save a lot if konsole would not link to libkio. Does the delayed initialization of a lot of stuff in konsole actually still work ? (4 years ago on a K6/200 with KDE 2.x the same command took 1.9 seconds)

What numbers do you get ?

Happy new year
Alex

P.S. some days ago I had to compile the cdrtools written by Joerg Schilling from sources. I downloaded cdrtools sources and tried to build them. They shouted in my face that GNU make is broken and I should install smake written by Joerg Schilling. So I did this. As if we would need another make implementation...
It was installed in /opt/schily/. Hmm. And the readme complained that much to little software installs itself correctly under /opt/. But smake and cdrecord installed correctly under /opt/schily/. During make install it complained again that GNU make is broken and unmaintained and that the nice smake should be used instead.
It didn't make a good impression to me at all. Felt like shouting "I am the only one who does things right". Hmm...

Comments

Konsole is usually launched either
- via kdeinit
- directly when KDE is prelinked
You've measured something that doesn't matter much.


By Lubos Lunak at Fri, 12/30/2005 - 21:00

Try
time kwrapper konsole -e sh -c exit
then ;-)


Once you replace the sleep(1) in wrapper.c with a small nanosleep it may actually be useful.


By Waldo Bastian at Sat, 12/31/2005 - 01:28

> ... written by Joerg Schilling ...

He recently started a complete distribution of OpenSolaris:

And if you're looking for a flamewar then just say something negative about Solaris in the heise.de forum. (No, I didn't try it myself. ;-) )


By christian mueller at Fri, 12/30/2005 - 21:32

hmmm. do you know how to accurately measure this? i'd love to see some serious comparisons between several apps... comparing apps against each other in performance is quite difficult under linux.


By superstoned at Fri, 12/30/2005 - 22:19

Got 4.799s on the MacMini with KUbuntu 5.10 and kde 3.4.3... :S


By apol at Sat, 12/31/2005 - 01:10

I tried it on Debian SID with xorg 6.8.2 and it said 1.0 more or less but I've upgraded and now it tells 0m0.796s, is it possible? :S


By apol at Sat, 12/31/2005 - 02:53

HP OmniBook xt1000, 1.2GHz, 256mb RAM.

real 0m6.944s
user 0m1.481s
sys 0m0.227s

That is while qt-copy is compiling, though ;-)


By cniehaus at Sat, 12/31/2005 - 10:02

Values in milliseconds with 10 samples each...
Konsole: mean 549.2, stdev 4.21
XTerm: mean 104.6, stdev 4.65

Not to mention, Konsole uses around 75% of that time on the CPU, and XTerm's highest usage was 24%.


By sapphirecat at Sat, 12/31/2005 - 11:32