kde4 doesn't utilize mutiple available CPU cores

[Kevin Hunter]

Note: This bug report may be misplaced because I'm not sure to which package/program to attribute the oversight/problem. Is it a kernel package I'm perhaps missing? Nothing immediately jumped out at me while perusing the kernel packages via Adept.

Setup:
Dell Vostro 1500, Core 2 Duo
Installed Hardy alpha 6, KDE4

To save someone 5 minutes of head scratching, I'll repeat: that's KDE4, not KDE3. However, looking through the /etc/init.d/kdm script, I see references to kde3, so not sure ...

Background: I have a Core 2 Duo at 2.2 GHz, with plenty of RAM. That means effectively 2 CPUs or cores. While researching/preparing https://bugs.launchpad.net/ubuntu/+source/kdebase-workspace/+bug/204136 , my kdm binary started hogging a core. This still left a free available core. However, things noticeably slowed down. For instance, typing the previous bug report in Konqueror was slow. On a single core system I'd expect this, but on a dual-core system, with one core not being utilized, I would expect to not immediately notice a hogging process like that.

Symptom: KDE4 responsiveness gets noticeably slower when a process (in this case, kdm) hogs a core, despite the fact that a second core is barely being used.

Steps to reproduce:
0. Use a multi-core processor. In this case, Intel based.
1. Login to a KDE4 session.
2. Get a process to go hog-wild and take 100% of a core. (In this case, kdm, but I'm sure a busy wait loop will do the trick.)
3. Use the system.

Actual behavior:
General system slow down, becomes slow to respond to my inputs and to draw the screen.

Expected behavior:
Hog-wild broken behavior of kdm aside, I expect to notice a slow down if there's an unused core. Programs ought to use the unused core.

[Yuriy Kozlov (yuriy-kozlov)]

Thank you for filing this bug. Please check if you are running a kernel that supports multicore. Post the output of
uname -a

[Kevin Hunter (hunteke)]

$ uname -a
Linux hani-alpha 2.6.24-12-generic #1 SMP Wed Mar 12 22:31:43 UTC 2008 x86+64 GNU/Linux

[Kevin Hunter (hunteke)]

$ uname -a
Linux hani-alpha 2.6.24-12-generic #1 SMP Wed Mar 12 22:31:43 UTC 2008 x86_64 GNU/Linux

[Kevin Hunter (hunteke)]

> Expected behavior:
> Hog-wild broken behavior of kdm aside, I expect to
> notice a slow down if there's an unused core. Programs
> ought to use the unused core.

Apparently I was tired and typoed my grammar. Should be

"... I expect *not* to immediately notice a slow down if there's an unused core. ..."

[Bleh7777 (bleh777777777777+launchpad)]

I've got the same problem with Hardy RC, running GNOME. Hardware is an AMD Athlon X2 5600+ with 3 GB RAM.

$ uname -a
Linux Acer 2.6.24-16-generic #1 SMP Thu Apr 10 13:23:42 UTC 2008 i686 GNU/Linux

One thing I did notice though, is that both cores are utilized when I set concurrency to 'shell' in /etc/init.d/rc, but that causes HAL to stop working.

[Yuriy Kozlov (yuriy-kozlov)]

This needs to be tested with:
1. Vanilla Ubuntu
2. No GUI, just a couple cpu intensive processes

Possible kernel or init bug?

[Kevin Hunter (hunteke)]

Not to pressure the awesome folks actually doing the work, but how does this bug move from incomplete to another status? I've seen more than a few bugs get pigeonholed into the incomplete status and then just sit there.

[Yuriy Kozlov (yuriy-kozlov)]

Kevin, as I mentioned above, one of the things that will help to narrow this down is to test it with different configurations, particlarly other desktop environments than KDE4 to see if it's a lower level issue or something specific to KDE.

[Kevin Hunter (hunteke)]

Cool, thanks. Just checking in.

[Kevin Hunter (hunteke)]

For #1: For some anecdotal evidence, since Hardy came out, I've been using the stock version of that on this same hardware. I haven't noticed anything directly, but then I haven't been trying to test it out as thoroughly.

For #2: Would this work as a cpu-intensive process?
---- test.c ----
int main ( ) {
 while ( 1 );
}
----
$ gcc -s -o test test.c

[Kevin Hunter (hunteke)]

Not pressuring, just nudging so this marked-as-incomplete bug won't time out.

[Yuriy Kozlov (yuriy-kozlov)]

FWIW, Intrepid is definitely utilizing both cores here.

[Harald Sitter (apachelogger)]

So, what to do?