CPUload = 1.00 and nothing running

destroyedlolo

Hello,

I have a BananaPI on which "nothing" is running : no apache, database, ... only standard daemons.

As per top, the system load is 1.00, and I didn't see anything running on it.

More, I got a 2nd similar Banana, on which I'm running Apache, PHP, PostgreSQL (a low traffic web site but with some activities), + DHCP, Named, ntpd, ... and it's load is b/w 0.1 and 0.4 when I got some visitor.

The only difference I can see, is the second one is under 3.4.90 whereas the other is under 3.4.103.

As I can't see anything on top ... how can see what is creating such load ?

Thanks

Laurent

Naib · Posted: Sun Dec 28, 2014 7:01 pm Post subject:

Loading in linux isn't like in windows, its an indication of threads waiting.
Now it might be due to computational time, but it also could be due to WAITING (usually I/O or a zombie)

Could you paste the 3 top lines of TOP
The 3rd will show you waiting percentage
_________________

destroyedlolo · Posted: Sun Dec 28, 2014 7:50 pm Post subject:

Hello,

As you can see, top does't highlight any waiting thread :

Naib · Posted: Sun Dec 28, 2014 8:03 pm Post subject:

two things

1) could you paste the complete output of: ps aux
2) do you have htop, its generally a lot clearer to see what is actually eating your CPU
_________________

destroyedlolo · Posted: Sun Dec 28, 2014 10:21 pm Post subject:

Naib · Posted: Sun Dec 28, 2014 11:39 pm Post subject:

ok you definitely do have something waiting & that would explain the load = 1 (while also have low CPU usage).

try: vmstat 1
for say... 20seconds and paste the output.
_________________

destroyedlolo · Posted: Mon Dec 29, 2014 12:23 am Post subject:

Naib · Posted: Mon Dec 29, 2014 1:55 am Post subject:

you seem to have some flushing going on. That might be part of the problem.

Two other things

1) dmesg. this could be spewing a load of ATA errors

2) look at ps aux | grep [f]lush & compare to

grep ^ /sys/class/block/*/dev

The number after flush will correspond to a block device, this will tell you what device (ram, disk...) is being flushed.

see if this is consistent.
_________________

destroyedlolo · Posted: Mon Dec 29, 2014 10:23 pm Post subject:

Hi,

I don't have any error message in system's log.

Naib · Posted: Mon Dec 29, 2014 10:39 pm Post subject:

That's what those [flush-...]. Kernel processes are.

From the additional information its associated with /dev/sda ( not the partition) and mmcblk0. So that explains why it isn't showing up on Inotify ( no actual files) its like the disk write request are too high for the pi to handle

Maybe your "dirty ratio" is not optimal for your environment.

cat /proc/sys/vm/dirty_ratio
cat /proc/sys/vm/dirty_background_ratio
_________________

destroyedlolo · Posted: Mon Dec 29, 2014 10:56 pm Post subject:

Naib · Posted: Mon Dec 29, 2014 11:06 pm Post subject:

Could you try, as root
echo "40" > /proc/sys/vm/dirty_ratio

And see what your loading is

--edit--
Have a read of this
https://lonesysadmin.net/2013/12/22/better-linux-disk-caching-performance-vm-dirty_ratio/
_________________

Ant P. · Watchman Joined: 18 Apr 2009 Posts: 6920

Naib · Posted: Tue Dec 30, 2014 12:06 am Post subject:

destroyedlolo · Posted: Tue Dec 30, 2014 8:45 pm Post subject:

The full name is [usb-hardware-sc].

Now the interesting thing is ... I didn't connected anything to USB port (and unlike the Raspberry, usb data flow is dedicated, no shared with the network for example).

I'm checking with the FEX file : as some A20's PINs are multiplexed, it may be a side effect ... checking ...

update : I found a thread related to that, https://groups.google.com/forum/#!topic/linux-sunxi/iNkoc1dI394
But it is quite old ...

destroyedlolo · Posted: Tue Dec 30, 2014 9:24 pm Post subject:

ok, I found it : it seems it's the OTG feature that is buggy.
Modifying the fex file as per this page.

Thanks for your help.
I'm reverting back to Gentoo forum to see push for a fix in the "official" kernel.