You are not logged in.

#1 2011-01-31 18:14:09

Sekre
Member
From: The Rainy North
Registered: 2006-11-24
Posts: 116

2.6.37 Still suffering from high load

Hello,

as litemotiv pointed out in this thread the load issue was indeed solved in 2.6.37 as can be seen here in this commit. And per suggestion I now create a new thread to investigate as to why I still have high load.

Here is the the info from 53 minutes uptime. About 40 of these were idle.

[braino] ~/ uname -a
  Linux braino 2.6.37-ARCH #1 SMP PREEMPT Sat Jan 29 20:00:33 CET 2011 x86_64 Intel(R) Core(TM) i5 CPU M 430 @ 2.27GHz GenuineIntel GNU/Linux
[braino] ~/ uptime
  18:39:20 up 53 min,  2 users,  load average: 1.05, 1.12, 1.13

And here is the current where I have been playing music, chatting and browsing (basically been trying to simulate normal use)

[braino] ~/ uptime
  18:59:57 up  1:13,  2 users,  load average: 1.00, 1.07, 1.12

Now I don't know too much about this unfortunately but as far as I have understood. Shouldn't it be the other way around since my activities would have scheduled more jobs than an idle system would and thereby increasing load rather than lower it? Could this be related to frequency scaling for the cpu? I have not noticed any clear differences between performance and ondemand though.

Here is a cutout of powertop with the top 10 causes for wakeups.

Cn                Avg residency       P-states (frequencies)
C0 (cpu running)        ( 1.3%)       Turbo Mode     5.3%
polling           0.0ms ( 0.0%)         2.27 Ghz     0.0%
C1 mwait          3.1ms ( 7.6%)         2.14 Ghz     0.0%
C2 mwait          5.3ms (19.6%)         2.00 Ghz     0.0%
C3 mwait          6.1ms (71.5%)         1199 Mhz    94.7%

Wakeups-from-idle per second : 179.0    interval: 15.0s
no ACPI power usage estimate available

Top causes for wakeups:
  42.8% ( 72.6)   PS/2 keyboard/mouse/touchpad interrupt
  10.1% ( 17.2)   kworker/0:1
   8.2% ( 13.9)   [iwlagn] <interrupt>
   7.0% ( 11.8)   [hda_intel, mmc0, nvidia] <interrupt>
   6.4% ( 10.9)   [ahci] <interrupt>
   4.0% (  6.7)   kworker/0:0
   3.5% (  6.0)   [kernel scheduler] Load balancing tick
   3.1% (  5.3)   [hda_intel] <interrupt>
   3.1% (  5.3)   alsa-sink
   3.0% (  5.0)   syndaemon

Other than this I don't really know what to post here so I will leave it at this now and await questions for more information.
If it makes difference this is on a laptop (Lenovo Thinkpad 410i) running an core i5 with two cores and two hyper-threads.

Grateful for any advice on where to investigate or possible causes for this! smile

Offline

#2 2011-01-31 19:00:27

ja1217
Member
Registered: 2007-12-29
Posts: 13

Re: 2.6.37 Still suffering from high load

This sounds very similar to the issues I had been having since .35.   In .35, there were kslowd processes that would pop up and consume just about all of the CPU.  In .36 these changed to the "kworker" processes.  From what I understand, the high load comes from device polling and mainly seems to effect Thinkpad's for some reason.  I was able to fix the issue by adding

options drm_kms_helper poll=N

to /etc/modprobe.d/modprobe.conf and reboot.  This disables the polling (I think).  Since I have done this, I have had normally load/cpu and everything works great and I haven't noticed any downsides to disabling the polling.

Hope this helps.

Last edited by ja1217 (2011-01-31 19:03:32)

Offline

#3 2011-01-31 19:19:10

fowler
Member
Registered: 2009-01-26
Posts: 152

Re: 2.6.37 Still suffering from high load

I just made that change and it didn't help.  It also affects Dell laptops.

Offline

#4 2011-01-31 19:32:35

litemotiv
Forum Fellow
Registered: 2008-08-01
Posts: 5,026

Re: 2.6.37 Still suffering from high load

fowler wrote:

I just made that change and it didn't help.  It also affects Dell laptops.

Please don't hijack the topic fowler. ja1217 stated that it was a fix targeted towards Thinkpads, which the topic starter is also using.


ᶘ ᵒᴥᵒᶅ

Offline

#5 2011-01-31 19:42:30

fowler
Member
Registered: 2009-01-26
Posts: 152

Re: 2.6.37 Still suffering from high load

Sorry, but I'm not trying to hijack this thread.  I guess I should have said that I also have this issue that affects both my Dell AND my Thinkpad and that the proposed fix didn't work for me on both laptops.

Offline

#6 2011-01-31 19:43:13

Sekre
Member
From: The Rainy North
Registered: 2006-11-24
Posts: 116

Re: 2.6.37 Still suffering from high load

ja1217 wrote:

This sounds very similar to the issues I had been having since .35.   In .35, there were kslowd processes that would pop up and consume just about all of the CPU.  In .36 these changed to the "kworker" processes.  From what I understand, the high load comes from device polling and mainly seems to effect Thinkpad's for some reason.  I was able to fix the issue by adding

options drm_kms_helper poll=N

to /etc/modprobe.d/modprobe.conf and reboot.  This disables the polling (I think).  Since I have done this, I have had normally load/cpu and everything works great and I haven't noticed any downsides to disabling the polling.

Hope this helps.

tried this now, but unfortunately I see no load difference at all actually.

Though I now have a slightly longer log (looked like this every boot since 2.6.35-ish)

 20:27:20 up 1 min,  1 user,  load average: 2.78, 0.77, 0.26
 20:29:06 up 2 min,  1 user,  load average: 2.07, 1.19, 0.48
 20:33:02 up 6 min,  1 user,  load average: 1.09, 1.16, 0.62
 20:34:34 up 8 min,  1 user,  load average: 1.02, 1.12, 0.66
 20:41:32 up 15 min,  1 user,  load average: 1.00, 1.08, 0.82

It seems to settle after about 5-6 minutes after boot (same with or without starting X) and then normalizes to somewhere between 1 and 1.5

Offline

#7 2011-01-31 20:01:14

ja1217
Member
Registered: 2007-12-29
Posts: 13

Re: 2.6.37 Still suffering from high load

My hardware is older (x200, core2duo, 4500mhd for video) so that may be why it worked for me and not you.  It must be a different problem that causes very similar symptons then, so I don't know.

Good luck.

Offline

#8 2011-01-31 20:13:01

litemotiv
Forum Fellow
Registered: 2008-08-01
Posts: 5,026

Re: 2.6.37 Still suffering from high load

fowler wrote:

Sorry, but I'm not trying to hijack this thread.  I guess I should have said that I also have this issue that affects both my Dell AND my Thinkpad and that the proposed fix didn't work for me on both laptops.

Ok thanks for the explanation fowler, carry on. wink


ᶘ ᵒᴥᵒᶅ

Offline

#9 2011-02-02 22:40:00

hvonbank
Member
Registered: 2011-02-02
Posts: 2

Re: 2.6.37 Still suffering from high load

I have what I think is this same issue with a Thinkpad T61.

It initially appeared a month or so ago when I did an update.  After trying all of the fixes I saw on google without any luck, I downgraded intel-dri, libgl, and xf86-video-intel from the version in testing to what was in extra at the time, and that took care of it.  I think I tried this after changing switch to use the vesa drivers took care of it.  I just assumed it was a bug in the new package in testing that would get corrected upstream.

However, after running an update last night, this issue reappeared when I rebooted my system today.  I normally don't shutdown or reboot my machine very often, so it is possible it is an update in the last couple weeks that is causing the problem.

I just tried downgrading the kernel, the packages I listed above, and a few other things to older packages, but this time it did not take care of the issue.  I have not repeated all of the steps I listed above yet, but I'm still assuming this bug is related to an X.org driver or similar.

It is making my computer almost unusable, as if I type too quickly on the keyboard some of the keystrokes are getting lost.

Note that I am running Arch 64-bit and XFCE as my window manager.

Update: The issue occurring now does not appear to be related X or video drivers as it appears without X starting or i915 modules being loaded.

Last edited by hvonbank (2011-02-04 17:57:11)

Offline

#10 2011-02-03 01:36:34

Sekre
Member
From: The Rainy North
Registered: 2006-11-24
Posts: 116

Re: 2.6.37 Still suffering from high load

hvonbank wrote:

I have what I think is this same issue with a Thinkpad T61.

It initially appeared a month or so ago when I did an update.  After trying all of the fixes I saw on google without any luck, I downgraded intel-dri, libgl, and xf86-video-intel from the version in testing to what was in extra at the time, and that took care of it.  I think I tried this after changing switch to use the vesa drivers took care of it.  I just assumed it was a bug in the new package in testing that would get corrected upstream.

However, after running an update last night, this issue reappeared when I rebooted my system today.  I normally don't shutdown or reboot my machine very often, so it is possible it is an update in the last couple weeks that is causing the problem.

I just tried downgrading the kernel, the packages I listed above, and a few other things to older packages, but this time it did not take care of the issue.  I have not repeated all of the steps I listed above yet, but I'm still assuming this bug is related to an X.org driver or similar.

It is making my computer almost unusable, as if I type too quickly on the keyboard some of the keystrokes are getting lost.

Note that I am running Arch 64-bit and XFCE as my window manager.

Ah, that's a bummer dude hmm

We could indeed share a similar problem. Though at least for me this does not seem to be a problem with X since I experience high loads even without X. And although obscure in my powertop output I don't have an intel card in mine but an nvidia card. Which is pretty much the reason I couldn't downgrade due to instability with tv-out on my chipset...

You say you have trouble if you type too fast, while I don't have that particular problem for me it causes problems when running intensive tasks such as playing full hd videos. Doing two of these tasks at once or something else that generates a lot of jobs can cause playback to fail or glitch even though usage is still low on the cpu.

Edit: oh, also welcome to the forums tongue

Last edited by Sekre (2011-02-03 01:37:23)

Offline

#11 2011-02-03 04:53:06

hvonbank
Member
Registered: 2011-02-02
Posts: 2

Re: 2.6.37 Still suffering from high load

The issue I had a month ago may have been different, because it is also occurring for me now when I don't load X.  I am also seeing almost exactly the same output from powertop as you are.

While I haven't tracked down the cause of this yet, I was able to temporarily fix it by disable auto module loading in rc.conf, and just manually adding the modules I need to MODULES=().  I'll see if I can narrow it down to a particular kernel module that is causing the problem. 

-Henry

Offline

#12 2011-02-03 13:37:54

helloworld1
Member
Registered: 2010-12-26
Posts: 60

Re: 2.6.37 Still suffering from high load

The same for me in Thinkpad x201 core i7, intel GMA HD graphics.

Offline

#13 2011-02-04 06:54:05

Kirurgs
Member
Registered: 2008-10-20
Posts: 141

Re: 2.6.37 Still suffering from high load

Hi!

I have T500, 2.6.37 + catalyst... I have disabled DRM stuff in MODULES and it won't load, also I can't see any DRM stuff doing lsmod | grep drm, so this is not an issue with DRM at all. I assume high wakeups are a result of kernel or upower/udisks or so. I have no high load average, just wakeups make me worried...
This is a shot from my current situation, nothing heavy is running background:

top - 08:45:17 up 3 days, 23:48,  3 users,  load average: 0.15, 0.07, 0.11
Tasks: 233 total,   1 running, 230 sleeping,   0 stopped,   2 zombie
Cpu(s): 13.3%us, 10.2%sy,  0.1%ni, 75.6%id,  0.7%wa,  0.1%hi,  0.1%si,  0.0%st
Mem:   8115472k total,  7666636k used,   448836k free,   107744k buffers
Swap:  6834204k total,        0k used,  6834204k free,  1454712k cached

I'm surprised about one thing (see powertop output below), I see that I have total of 3.2K wakeups and if I count them in the list (about 1K), the sum is nowhere near actual wakeups shown (3.2K), what is that? Even if we count like the rest of the processes below clock-applet with wakeup 1, this adds roughly 200 to wakeups, where those 3K come from?

Cn                Avg residency       P-states (frequencies)
C0 (cpu running)        (25.1%)       Turbo Mode     0.0%
C0                0.0ms ( 0.0%)         2.54 Ghz     0.0%
C1 mwait          0.0ms ( 1.2%)         1.60 Ghz    50.0%
C2 mwait          0.1ms (12.1%)          800 Mhz    50.0%
C3 mwait          0.4ms (61.6%)

Wakeups-from-idle per second : 3432.8   interval: 10.0s
no ACPI power usage estimate available

Top causes for wakeups:
  13.6% (238.8)   kworker/0:0
   9.3% (162.8)   java
   8.6% (150.9)   [kernel scheduler] Load balancing tick
   7.1% (124.7)   firefox
   5.8% (102.1)   wineserver
   5.4% ( 94.0)   [hda_intel] <interrupt>
   5.2% ( 90.5)   firefox-bin
   2.9% ( 50.0)   [kernel core] hdaps_mousedev_poll (hdaps_mousedev_poll)
   2.8% ( 48.5)   [ahci] <interrupt>
   1.2% ( 20.3)   wnck-applet
   1.1% ( 19.9)   frmbld.EXE
   0.7% ( 12.8)   skype
   0.7% ( 12.2)   evince-previewe
   0.6% ( 10.1)   geany
   0.6% ( 10.0)   gnome-hdaps-app
   0.3% (  5.0)   [eth0] <interrupt>
   0.3% (  5.0)   syndaemon
   0.3% (  5.0)   gnome-do
   0.3% (  4.8)   [acpi] <interrupt>
   0.3% (  4.5)   gnome-netstatus
   0.2% (  2.8)   nautilus
   0.2% (  2.8)   soffice.bin
   0.1% (  2.4)   [kernel core] hrtimer_start (tick_sched_timer)
   0.1% (  2.0)   netspeed_applet
   0.1% (  1.8)   python2
   0.1% (  1.0)   clock-applet

When idling I also noticed that wakeups are too high, doing nothing only gnome session is opened it consumes about 150-200 or so. I searched google and nothing relevant pop up, so currently I'm just working as it is... This all results in short battery life, currently I have like 9c battery and 75Wh left of it and that lasts about 2-2.5hrs which is normal for a lot smaller battery not this big one!
I hope there is or will be a cure for this shortly...

Last edited by Kirurgs (2011-02-04 06:59:24)

Offline

#14 2011-02-04 14:48:28

fowler
Member
Registered: 2009-01-26
Posts: 152

Re: 2.6.37 Still suffering from high load

I too have a Thinkpad x201 core i7, intel GMA HD graphics.  I noticed this problem when my battery life essentially was cut in half from 4 to 2 hours.

Offline

#15 2011-02-06 21:38:22

krum
Member
From: Grenoble - FR
Registered: 2008-04-10
Posts: 32

Re: 2.6.37 Still suffering from high load

Same issue on a HP 8440p with i5 CPU.
No X, minimal install, doing nothing and a perfect 1 of load.

Maybe just a bug calculating the load, not a real load of 1.

Offline

#16 2011-02-08 23:37:06

ecki
Member
From: Germany
Registered: 2011-02-08
Posts: 2

Re: 2.6.37 Still suffering from high load

I think this load could be coming from the intel_ips module. I just unloaded the module and my load droped down to normal.

While searching for an answer I discovered another high load problem (though unrelated) in the linux kernel mailing list and followed a step from an answer at http://kerneltrap.org/mailarchive/linux … 0/2/327817

When I issued "ps ax | grep ' D'" I always saw the kernel thread [ips-monitor] in D state. After some more googling I found that this thread might have been introduced with a IPS patch (intelligent power sharing) for Intel chips (intel_ips).

I don't know what the implications are from removing the intel_ips module, but for the last few minutes at least my laptop survived big_smile If I understand this correctly the module is responsible for potentially raising performance of the GPU as long as the thermal state of the system has some headroom. I don't need too much GPU performance, so for me this doesn't seem to be an issue.

Offline

#17 2011-02-09 12:39:27

pogeymanz
Member
Registered: 2008-03-11
Posts: 1,020

Re: 2.6.37 Still suffering from high load

I can confirm that unloading the intel_ips module decreases the load average back to normal!

Unfortunately this doesn't completely alleviate the high interrupts reported by powertop, though it does seem to help.

Offline

#18 2011-02-10 01:19:02

ffjia
Member
Registered: 2009-08-26
Posts: 94

Re: 2.6.37 Still suffering from high load

Thanks ecki for the workaround smile It works!

Is this a kernel bug or something?

Offline

#19 2011-02-10 16:46:52

ViruSzZ
Member
From: The Streets
Registered: 2010-10-14
Posts: 202
Website

Re: 2.6.37 Still suffering from high load

I can also confirm that after I unload the 'intel_ips' module the load decreased from 1.5 to 0.18 with the same apps opened.

Out


Their Momma Made Em, Their Momma Gave em & now she can`t even SAVE`em | My WebLog

Offline

#20 2011-02-10 17:26:44

unknwn
Member
Registered: 2008-02-02
Posts: 34

Re: 2.6.37 Still suffering from high load

Same here. After unloading intel_ips module load decreases load back to normal.

Offline

#21 2011-02-10 19:44:57

ecki
Member
From: Germany
Registered: 2011-02-08
Posts: 2

Re: 2.6.37 Still suffering from high load

ffjia wrote:

Is this a kernel bug or something?

I compiled a vanilla kernel with some thinkpad specific drivers and intel_ips as the only module. And again with the intel_ips module the load raises, without it drops. I also booted a grml 2010.12 and could verify that they have the same symptoms. So I think it is save to say that this is not arch specific.

Has anyone experience with filing kernel bug reports?

Offline

#22 2011-02-17 13:38:48

Sekre
Member
From: The Rainy North
Registered: 2006-11-24
Posts: 116

Re: 2.6.37 Still suffering from high load

Hello, terribly sorry for not answering here earlier been trying to specify and get my thesis accepted so I can start to work on it.

But yes, I can verify as well that the intel_ips module is the cause to the high loads on my system as well. Thanks for the find, I would never have dared to unload that one since I don't know what it does tongue

Offline

#23 2011-02-24 12:12:44

elstupidos
Arch Linux f@h Team Member
From: Santa Barbara, CA
Registered: 2008-01-14
Posts: 43

Re: 2.6.37 Still suffering from high load

I too was being plagued by high loads with the 2.6.36/37 kernels. Removing intel_ips worked for me as well and brought my load avgs to normal. Toshiba laptop core i3 cpu.

Offline

#24 2011-02-24 12:44:52

Ultraman
Member
Registered: 2009-12-24
Posts: 242

Re: 2.6.37 Still suffering from high load

I'm still experiencing this at the moment. Details can be found here.
But it only seems to affect me when running on battery power.

I haven't tried turning off the drm_kms_helper poll option, so I will try that this evening. Hope that helps. I will report back.

If I should open my own thread, tell me and I will. But I feel this might be related.

Offline

#25 2011-02-28 13:28:36

Telkkar
Member
From: United States
Registered: 2010-12-26
Posts: 38

Re: 2.6.37 Still suffering from high load

Ultraman wrote:

I'm still experiencing this at the moment. Details can be found here.
But it only seems to affect me when running on battery power.

I haven't tried turning off the drm_kms_helper poll option, so I will try that this evening. Hope that helps. I will report back.

If I should open my own thread, tell me and I will. But I feel this might be related.

Given that I'm the author of the linked thread, I thought I'd throw a little bit more into the fire.

I'm still having the problems, even though I posted mine as being fixed in the other thread (need to fix that). High interrupts while sitting idle, on either AC or battery.

Offline

Board footer

Powered by FluxBB