Troubleshooting a high load average on idle.

Ypnose · 2012-05-23 10:15:38

As Butcher said before, never had more than 1.0 with my Phenom II X4.
I'm using dwm + full CLI + firefox. It's very minimal, yesterday I had 1.20 1.14 0.98 just with mumble and firefox (3.3.6).
In idle (disabled everything), watched the load average and it was surprisingly high. With 3.3.4 (solved the issue) I have 0.01 0.03 0.03 in idle.

/var/log/Xorg.0.log seems to be fine, nothing changed. My specs are in my signature but it's AMD + NVIDIA.

Last edited by Ypnose (2012-05-23 10:15:56)

lagagnon · 2012-06-07 15:23:32

The high system load fault still exists with Kernel 3.3.7-1-ARCH

Shyne · 2012-06-17 06:30:27

Same problem here with kernel 3.4.2-2.

Boohbah · 2012-06-18 10:05:12

Same problem with 3.4.2-2-ARCH. Switching to 3.4.2-3-ck fixed it.
It appears to be scheduler related.
Here's a similar Ubuntu bug:
https://bugs.launchpad.net/ubuntu/+sour … bug/985661

loofygun · 2012-06-22 11:51:09

Same with 3.4.3-1-ARCH. I'm running a headless server, and am experiencing very high loads.

Edit. Just adding that I'm running 32bit Arch.

Last edited by loofygun (2012-06-25 14:07:42)

karol · 2012-06-22 18:00:43

I'm running linux 3.4.3-1 on a 32-bit Arch and thing's are fine here.

skunktrader · 2012-06-22 23:14:46

Perhaps its an 64-bit thing then. I am running 3.4.3-1 here on x86_64 and things are NOT fine here

esquiso · 2012-07-01 23:14:09

I'm runing linux-3.4.4-2 and I still get high load even on idle. Can someone confirm that, if X is not running, the idle load is much lower? I'm getting >0.7 idle on X and ~0.2 idle on a tty. Nonetheless, I was averaging 0.02~0.05 a few kernel versions ago (3.3.4, IIRC, was the last version where my loads were "normal" - i.e., low on idle, being on X or not).

#edit

Running linux-lts-3.0.36-1, load is back to normal.

Last edited by esquiso (2012-07-02 14:35:36)

karol · 2012-07-02 17:08:35

linux 3.4.4-2 on a 32-bit Arch, no load-related issues neither in X nor in tty.

karol · 2012-07-02 18:05:58

@esquiso
A bit far fetched, but maybe http://www.h-online.com/open/news/item/ … 29805.html -> https://lkml.org/lkml/2012/6/30/122 has something to do with it.
Did you test it with base X or were you running e.g. a web browser?

codekoala · 2012-07-03 06:37:18

I've seen abnormally high load averages on my work machine for quite a while now. It happened over the weekend, and it's happening again right now. I can't even SSH into it to get the output of uname, but it's running the 64-bit 3.4.3 kernel on a 3.4Ghz i7 with 16GB RAM. I avoid upgrading the kernel so I don't have to bother with recompiling VMware Workstation garbage for those few occasions when I can't use the ESX server.

Daily load average Weekly load average

I find that Ctrl+Alt+BS'ing X brings my load average back to normal. This is an acceptable solution for not losing my screen/tmux sessions, but it's certainly less than ideal. I would love to find an answer to this problem.

Other than those "spikes", you can see how my load average is relatively high the rest of the time even though my system is mostly idle except for gVim and a bit of surfing during the day.

lagagnon · 2012-07-04 06:01:09

I am not sure if I am on the right track or not but I am beginning to suspect either X and/or nVidia rather than the kernel. Why? I run wmii window manager and I have a status bar that shows me CPU Freq for both CPU cores. For over a month now one or other core bumps up from 800 to 2800MHz very quickly on about a 5-7 second interval. When I dropped out of X into runlevel 3 or ran fluxbox instead of wmii the CPU Freq stayed low.

That led me to suspect my status bar code, so I took it apart and I discovered the culprit was the code to show GPU Temp:

nvidia-settings -q=GPUCoreTemp

When I removed that code my CPU Freq stayed low at 800MHz.

Also, I now note that my load averages seem to be back to normal. That is why I am now beginning to suspect either the nVidia module or X - why should simply polling the GPU for its temperature cause the demand for maximum CPU speed and apparent increases in load?

Roken · 2012-07-04 09:50:25

I have the same issue with 3.4 series of kernels. ATI Gfx/AMD CPU, so we can rule out the whole Intel/nVidia theories.

My system is fine with 3.2 kernel, so I'd suggest that the finger is well and truly pointed towards the kernel.

Boohbah · 2012-07-07 19:13:36

More information from the ubuntu bug:
https://bugs.launchpad.net/ubuntu/+sour … omments/70
https://bugs.launchpad.net/ubuntu/+sour … omments/76

It should be fixed in 3.5

Roken · 2012-07-10 01:20:08

Oddly, after having no problems since dropping to linux-ice (3.2.13-2) I decided to rebuild the linux-ice kernel with ck/bfs patches. After a reboot load averages were sky high again (I waited a good ten minutes leaving the system at idle to let it settle down).

Built back without the patches and load average is normal again. This would suggest that there's also something in those patches contributing to this.

Darkimmortal · 2012-07-11 20:11:34

I've seen this bug on my server running x64 in every kernel since 3.3.4-2. Right now I'm seeing load averages of 12 when normally it would be 0.5-1.5.

It seems to be having a nasty effect on realtime stuff (SRCDS etc.) so I will be reverting again

Boohbah · 2012-07-17 03:41:38

Patch is in 3.5-rc7!

http://marc.info/?l=linux-kernel&m=134231428931763

Peter Zijlstra (1):
sched/nohz: Rewrite and fix load-avg computation -- again

donniezazen · 2012-07-17 06:31:47

@Boohbah It does lower system load but it doesn't fix higher power consumption and temperature.

Roken · 2012-07-17 07:08:00

rc7 does not fix the problem. I am still seeing high load average even at idle.

Boohbah · 2012-07-19 23:22:37

I am seeing decreased load averages with 3.5-rc7 when the system is lightly loaded. This patch fixes the load average calculation issue.

donniezazen wrote:

@Boohbah It does lower system load but it doesn't fix higher power consumption and temperature.

donnie: How are you measuring power consumption and temperature between kernel versions? Care to provide some numbers? This may be a separate issue.

donniezazen · 2012-07-20 03:09:50

powertop and sensors give you idea and then your fan runs at 4xxx. Right now my system is running 3.3 kernel at 10W and 49C, on 3.4 or 3.5 kernel it's 20W and temp is around 55-65C.

Load average has gone down so low to be true.

I haven't had much time to look into a utility which would graph these results to provide some scientific numbers.

Boohbah · 2012-07-21 00:59:05

Also fixed in 3.4.6. Carry on.

Arch Linux

#26 2012-05-23 10:15:38

Re: Troubleshooting a high load average on idle.

#27 2012-06-07 15:23:32

Re: Troubleshooting a high load average on idle.

#28 2012-06-17 06:30:27

Re: Troubleshooting a high load average on idle.

#29 2012-06-18 10:05:12

Re: Troubleshooting a high load average on idle.

#30 2012-06-22 11:51:09

Re: Troubleshooting a high load average on idle.

#31 2012-06-22 18:00:43

Re: Troubleshooting a high load average on idle.

#32 2012-06-22 23:14:46

Re: Troubleshooting a high load average on idle.

#33 2012-07-01 23:14:09

Re: Troubleshooting a high load average on idle.

#34 2012-07-02 17:08:35

Re: Troubleshooting a high load average on idle.

#35 2012-07-02 18:05:58

Re: Troubleshooting a high load average on idle.

#36 2012-07-03 06:37:18

Re: Troubleshooting a high load average on idle.

#37 2012-07-04 06:01:09

Re: Troubleshooting a high load average on idle.

#38 2012-07-04 09:50:25

Re: Troubleshooting a high load average on idle.

#39 2012-07-07 19:13:36

Re: Troubleshooting a high load average on idle.

#40 2012-07-10 01:20:08

Re: Troubleshooting a high load average on idle.

#41 2012-07-11 20:11:34

Re: Troubleshooting a high load average on idle.

#42 2012-07-17 03:41:38

Re: Troubleshooting a high load average on idle.

#43 2012-07-17 06:31:47

Re: Troubleshooting a high load average on idle.

#44 2012-07-17 07:08:00

Re: Troubleshooting a high load average on idle.

#45 2012-07-19 23:22:37

Re: Troubleshooting a high load average on idle.

#46 2012-07-20 03:09:50

Re: Troubleshooting a high load average on idle.

#47 2012-07-21 00:59:05

Re: Troubleshooting a high load average on idle.

Board footer