(Explained) irqbalance fails and power profiles missing

[MtnMan]

Hello all

My updated KDE 2011.6 is a real pleasure to use.  I do notice that heavy cpu load no longer alternates between cores like it did with the previous version.  I am 99% sure that irqbalance is responsible for this.  Looking in PCC under Manage system services, I see that it is checked to start on boot but says Stopped.  Clicking start does not make it run.  I could find nothing @ irqbalance.org that helps and another forum had info on how to stop it only.  

Hoping that someone can give me some hints on what to look at.  I have an Intel Core 2 Duo and cpufreq working ok.  Only other odd thing I see is that Power Management shows only 1 profile when it used to have several - would like to get that back the way it was but less important - just mentioned because it may be a clue?

[MtnMan]

Sorry to bump this but I am still looking for clues.

I burned another cd and re-installed with the same results.  Md5sum is good and the ISO was from a torrent.  Booting into the live cd the same conditions exist.  Is it possible to have a bad ISO with a good checksum?  Does everyone have only the performance power profile or is this a "random feature" that exists only on my ISO?  I realize that other profiles can be added but it just seems that something is different/wrong.

On the install if I try to restart or stop service irqbalance I get:
Stopping irqbalance:                                                 [FAILED]

Re-installing irqbalance from synaptic gives the same results.

Any clues/suggestions would be welcome.

[menotu]

Don't really know what to suggest - have you tried "unticking" the irqbalance switch in System Services - rebooting and trying to start it again, possibly from the command line.........

service irqbalance stop

and then

service irqbalance restart

[MtnMan]

Don't really know what to suggest - have you tried "unticking" the irqbalance switch in System Services - rebooting and trying to start it again, possibly from the command line.........

service irqbalance stop

and then

service irqbalance restart

Yes to all those.  Each time I am met with the response: Stopping irqbalance:                                                 [FAILED]
Except if I try with the alt+F2 there is no output - I guess that's what you mean by the command line.

Tried reinstall of irqbalance w/o effect.

I also tried get virtualbox but the error for that is: "Failed to fork new process: Exec format error" and "Failed to run getvirtualbox as user root".

Just seems like I'm getting too many "random features".  Downloading ISO again from mirror this time to burn cd#3 to see if results are the same.  BRB

BTW thanks for responding - I was thinking of changing the heading to something more interesting like Girls on trampolines to get more views.

[menotu]

I also tried get virtualbox but the error for that is:

I assume you mean getvirtualbox (no space) 

I also assume you've installed getvirtualbox (current version is 4.0.10) from Synaptic?

Did a quick web search and found an irqbalance bug report had been filed at Red Hat but have no idea if it's relevant to your problem

https://rhn.redhat.com/errata/RHBA-2011-0804.html

[AS]

Hi,

As far as I know, irqbalance act as a daemon when multiple CPU are installed, instead perform a "static balance" once upon each boot on multi core systems, then exit.

From this point of view, service irqbalance stop, is failing because effectively irqbalance is not running, from the same point of view this is normal.

MtnMan,
what kernel are you running ?

AS

[MtnMan]

I also tried get virtualbox but the error for that is:

I assume you mean getvirtualbox (no space)  

I also assume you've installed getvirtualbox (current version is 4.0.10) from Synaptic?

Did a quick web search and found an irqbalance bug report had been filed at Red Hat but have no idea if it's relevant to your problem

https://rhn.redhat.com/errata/RHBA-2011-0804.html

Yes - getvirtualbox from synaptic - sorry.  The synaptic version is for 4.0.10 but the current VB version is 4.0.12.  I did it the usual synaptic way - once Getvirtualbox had a menu entry I clicked that and that is when I had the error noted above.  

Since Tex said it was ok to d/l from the site I tried that next (picking the one at the bottom which is .run) had the effect of locking things up for about 15 mins. Yet still no VB.  Next tried logging in to root account and (menu) GetVirtualbox then worked but this isn't the way it normally has to be done right?

I looked at the RH link but couldn't tell anything from it.

Hi,

As far as I know, irqbalance act as a daemon when multiple CPU are installed, instead perform a "static balance" once upon each boot on multi core systems, then exit.

From this point of view, service irqbalance stop, is failing because effectively irqbalance is not running, from the same point of view this is normal.

MtnMan,
what kernel are you running ?

AS

Hi AS,
Kernel is stock on the latest cd - 2.6.38.8-pclos1.bfs

I don't know if I can agree with your statement about irqbalance - I think it treats multi CPU/multi core the same.  This is how I test:

Code: [Select]

while true; do true; done
that locks up one core (or half CPU capacity in my case) and when monitoring I can see the activity rotate back and forth between cores every few seconds, effectively keeping one core from heating abnormally.  Now the load with one instance of that code is not rotating
between cores as it used to - not a good thing.

I could look over past posts but there were some updates awhile back that started to make things go nuts for me.  Don't know if I need to delete/add/update/rollback something or what.  I decided to wait for this cd and in the meantime installed other DE/OSs without unexpected issues; don't think I have a hardware problem.  I really, really want PCLOS KDE to work well for me like it used to  - my issues seem to be more than just irqbalance.  If there is any info I can give please ask.

Edit: Forgot to mention that I did another d/l, burn and install and all remains the same.

[AS]

Code: [Select]

while true; do true; done
   

OK, I think to understand what you mean:

the first image refer to your loop and kernel 2.6.33.7
the second refer to the same loop and kernel 2.6.38.8
the third refer to kernel 2.6.38.8 running a virtuabox guest, within the guest it's executing mylivecd.

Something has changed, but it seems related to kernel (or may be irqbalance is badly interacting with new kernel ? )
anyway, I never realized about this because ans you can see from the third image, my heavy process doesn't show the issue.

EDIT: virtualbox is running on a single core virtual CPU

AS

[MtnMan]

AS> thank you for that - it did not occur to me that a different kernel would give different results.  You gave me ideas to try:
1.  A slightly newer kernel than the one on the cd gives the same results.

2.  Remembering from a CK blog posted by menotu awhile back about a certain patch it was remarked that some would have fewer problems with the performance governor as opposed to ondemand.  I tried that and see a better load distribution but loose scaling.

3.  I also tried the latest kernel without the BFS patch but with no improvement.

If I roll back to the 2.6.33.7 kernel can I keep the 275 nvidia driver or does it have to go back to 270?  Wasn't there something else that was changed for that transition?

[AS]

AS> thank you for that - it did not occur to me that a different kernel would give different results.  You gave me ideas to try:
1.  A slightly newer kernel than the one on the cd gives the same results.

2.  Remembering from a CK blog posted by menotu awhile back about a certain patch it was remarked that some would have fewer problems with the performance governor as opposed to ondemand.  I tried that and see a better load distribution but loose scaling.

3.  I also tried the latest kernel without the BFS patch but with no improvement.

If I roll back to the 2.6.33.7 kernel can I keep the 275 nvidia driver or does it have to go back to 270?  Wasn't there something else that was changed for that transition?

All tests I've done were using the nvidia-275.xxx drivers, (my card is a nvidia 8600GS), you can safely install the 2.6.33.7, only take care about the fact that the default kernel will be the latest installed. Installing the 2.6.33.7 will not remove your current kernel, if something will go wrong you will be able to boot using the current kernel. (You probably know about, but because of your word roll-back ....)

A note for those that doesn't have used gkrellm: the graph in green refer to user space processing load, the red graph refer to kernel space processing load, as you can see virtualbox do extensive use of kernel system calls.

AS

[MtnMan]

AS>The first image you have is similar to behavior that I once had.  Expected that I would get that with a fresh install updated and loaded with that kernel.  That did not give me those results though. 

When I started having problems, as I remember, it did not coincide with a kernel change.   

If you (or anyone) has any other suggestions please don't hesitate...

[AS]

     

Another couple of images: the large of gkrellm graphs are 240 pixels = 240 secs = 4 minutes (the same as in previous post images)
The first image refer to the while true loop, the second to mylivecd in virtualbox guest.
As you can see the .bfs kernel, and specifically the process scheduler is responsible for the different behavior and not the irqbalance.

Your wrote before, that you were using a .bfs kernel, so ... I don't know what is going on in your installation. Try a different .bfs kernel.

EDIT: added the third image, refer to the same kernel but using "powersave" governor, see post below


AS

[MtnMan]

The images you have from Virtualbox are hard for me to see.  I have 2 for you:

1.  Represents the current situation - KDE 2011.6 kernel 2.6.38.8-1bfs.  Note the heavy use of CPU 2 and lack of profiles.





2.  Made with Full Monty 2011.4  kernel 2.6.38.3-11.bfs before updates.  Note fair distribution of load and presence of default profiles.



Prior to some updates that occurred sometime after May 2011 the behavior was always like #2.  I have tried different kernels both with and without the CK patches without success - from my experiences I think it is not related to kernel only.  I also don't think it is related to irqbalance only.  I suspect it is caused by something else in the configuration that changed - just wish I were smart enough to figure out what. 

[AS]

Your graphs come from systems using the same CPU ?

About profiles, yes, they has been reset/deleted upon a big kde update.

[MtnMan]

"Your graphs come from systems using the same CPU ?"

If I understand your question correctly - it is 1 CPU w/2 cores - Intel C2D E6550

"About profiles, yes, they has been reset/deleted upon a big kde update."

Okay - no big deal.

I think my computer problems may be a small minority here.  I cannot say with certainty, but it may be possible that the updates I had when problems started for me are in the list here.  If so I don't know how to verify:

http://www.pclinuxos.com/forum/index.php/topic,91220.msg764562.html#msg764562