ubuntu 14.10, kernel 3.16 is giving me a constant 1.00 load average

[chronniff]

This has had me scratching my head for a few days now. I recently bought an asus N550JK-DB74T, and installed ubuntu 14.10. For some reason it starts up and the load average climbs to 1.00 in about the first 5 minutes, and then never goes below that. I can see that all the cpus are idle, top tells me that, so does conky. When I run ubuntu 14.04 from a usb stick it does not behave this way, also if I run the kernel from 14.04 on the 14.10 system, it does not behave that way. I have tried disabling intel pstates, but that makes no difference. I have also tried installing newer kernels, all of which give me the strange 1.00 load average. So I believe it has to be a kernel bug, but I was wondering if anyone else has noticed this odd behavior. I'm trying to improve my battery life, but I'm actually not positive it is affecting it....Its hard to tell because battery life isn't this laptops strength. i have googled but not really found anyone posting bug reports that sound like this.

Thanks to anyone who took the time to read this, and I appreciate any insight,
Dave

[Doug S]

I think this thread might get better exposure on the "general help" forum.
There have been some odd things mentioned on some threads when thermald is involved. Are you running thermald? If yes, try without it, but watch your temperatures. When using the intel_pstate driver and when your load average is 1, but you do not think it should be, what do yo get for this (example from my computer):

doug@s15:~/temp$ sudo cat /sys/devices/system/cpu/cpu*/cpufreq/cpuinfo_cur_freq
2052750
2052750
1840250
1627750
1999625
1734000
2052750
1680875

By the way, I am running kernel 3.16.0-29 on my desktop 14.10 test computer (it is a VM) and the load average is 0.00.

[chronniff]

I think this thread might get better exposure on the "general help" forum.
There have been some odd things mentioned on some threads when thermald is involved. Are you running thermald? If yes, try without it, but watch your temperatures. When using the intel_pstate driver and when your load average is 1, but you do not think it should be, what do yo get for this (example from my computer):

doug@s15:~/temp$ sudo cat /sys/devices/system/cpu/cpu*/cpufreq/cpuinfo_cur_freq
2052750
2052750
1840250
1627750
1999625
1734000
2052750
1680875

By the way, I am running kernel 3.16.0-29 on my desktop 14.10 test computer (it is a VM) and the load average is 0.00.

Hey, thanks for responding. Yeah i wasn't sure where to post this, i'll give the general help forum. I too have ubuntu 14.10 with the same kernel on my desktop running with a normal load average, its an old nehalem i7. I only see the problem on my laptop which has a i7-4710HQ which is a haswell generation i7. I have tried disabling both intel_pstate and thermald with no change. Here is a readout of my cpu frequencies

sudo cat /sys/devices/system/cpu/cpu*/cpufreq/cpuinfo_cur_freq
1394433
800000
1596093
799902
1585058
799707
880859
1063085

and after 12 minutes of having my computer on my load averages are 1.17, 1.10, 0.71. If i give it more time they will all migrate to around 1.

thanks again for helping

[QIII]

Moved to General Help

[Doug S]

and after 12 minutes of having my computer on my load averages are 1.17, 1.10, 0.71. If i give it more time they will all migrate to around 1.

The first listed load average has a 1 minute time constant, the second has a 5 minute time constant, and third has a 15 minute time constant. So there will always be a lag between the 3. Reported load average can be difficult to extract real meaning from, because it doesn't take variable clock frequencies into account.

Anyway, 4 of your CPUs have clock frequencies well off the lowest value of 800 MHz, and so there does seem to be some load. I'm surprised you don't see anything with top. Are you willing to experiment? You might want to try Kernel 3.19RC4

Edit: Opps, I forgot you said you already tried newer kernels.

[chronniff]

yeah I would definitely be willing to experiment, I'm treating the installation as temporary until I figure out the issue, or if I can't, installing 14.04. Yeah I have tried newer kernels, and I have even used the 14.04 kernel, which does not give me this behavior, the only one I haven't tried is 3.15.

Really I have no dire need to use 14.10 over 14.04 that I'm aware of, but now I feel like I really want to figure it out, since it seems to make no sense

[chronniff]

when I have pstates and thermald disabled I get this for my cpu frequencies but still get the same load average

800000
800000
800000
800000
800000
800000
800000
800000

[Doug S]

when I have pstates and thermald disabled I get this for my cpu frequencies but still get the same load average

800000
800000
800000
800000
800000
800000
800000
800000

Yes, the load Verses frequency response curves are different for the acpi cpufreq driver and the intel_pstate driver. Also the cpufreq driver doesn't report actual frequencies, but reports requested frequencies.

Your issue is very interesting.

If you have been using generic kernels, perhaps try a low-latency one as a test (or the other way around).
Are you comfortable compiling the kernel? Kernel bi-section could be used to isolate the exact commit that introduces the problem.

[chronniff]

yeah, I have mostly been using generic kernels, have one custom installed right now so I'm comfortable compiling,although you might have to walk me through what a kernel bi-section would be. My custom kernel is preemptible, but I only raised the wake up frequency to 300. I'm not sure what else goes into the lowlatency kernel, but there was no change in behavior on that kernel.

Just a thought here, but does IO load go into the load average, could it be something with my ssd?

I probably won't get to it this weekend, but maybe if I find some time.....thanks again for all your help, I really appreciate it

[Doug S]

I'm not sure what else goes into the lowlatency kernel, but there was no change in behavior on that kernel.

O.K. it was just a thought. The only difference that I know of between generic and lowlatency is the 250 Hertz basic clock (jiffy) goes to 1000 Hz.

Just a thought here, but does IO load go into the load average

Yes. Here is a cut and paste from some notes I made a few years ago while working on incorrect reported load averages:

"My CPU is not very busy, but my load averages are high":
 In both Linux and Unix, load averages include pending processes.
 In linux, as opposed to unix, load averages include I/O waits, such as waiting for disk reads or writes, and pending processes.

Howerver, you should be able to observe I/O waits on your "top" display.

could it be something with my ssd?

I suppose.

so I'm comfortable compiling, although you might have to walk me through what a kernel bi-section would be.

Sure. But note that I pretty much only use the kernels from kernel.org for this type of stuff. In the end, if some upstream bug is found and to be reported, they prefer not distro specific kernels to be used. Also note, that I failed at kernel bi-section more times than I have succeeded.