strange output frpm /proc/stat

MrUmunhum · 11-05-2014, 07:40 PM

Hi group,
While testing a new program on my Acer Aspire laptop running Fedora 13, I notice strange output from /proc/stat. The laptop has an AMD64 chip with 2 CPUs, when I start a loop script

Code:

while true ;do usleep 100; done

the output of /proc/stat shows both CPUs very busy.
I verified this with gnome-system-monitor. My program shows this:

I have access to one other machine and that one seams to produce valid output.

As anyone else noticed this anomaly?

Thanks for your time.

grail · 11-06-2014, 02:38 AM

I am curious what you are basing your evaluation of 'valid output' on?

Also, how have you confirmed that it is the loop causing the CPU usage?

MrUmunhum · 11-06-2014, 02:56 PM

OK, after thinking about this, I found an answer that makes sense to me.

The loop I am using has a 'usleep 100' in it, that allows the task scheduler to switch the CPU/Processor and there by causing both of the CPUs to be used, putting a load on both of them. If I use this loop:

Code:

while true ;do :; done

only one CPU is used and that one is pegged at 100%.

Interesting in sight into the internals of an OS.

MrUmunhum · 11-06-2014, 03:52 PM

Not to beat a dead horse but the output plot shows another interesting observation.

The plot shows using both loops, the first major spike is using the usleep, the second major spike does not. Notice the higher system usage ( Blue ) on the first spike. The second spike shows very little system usage.

Don't know how useful this information is but it seams interesting to me.

grail · 11-06-2014, 06:54 PM

Not sure if just me, but I am unable to see either attachment. Please make sure it is one of the types provided otherwise it will not display.

MrUmunhum · 11-08-2014, 03:40 PM

Quote:

Originally Posted by grail

Not sure if just me, but I am unable to see either attachment. Please make sure it is one of the types provided otherwise it will not display.

The images are on my server at 64.123.13.3. Try these links:

http://64.124.13.3/Images/pfssg.png
http://64.124.13.3/Images/pfssg_2.png

The second one is very interesting. It shows the outrages system overhead. More overhead than real work.

I ran this again on a 4 CPU machine with almost no real work being done. Here is the plot for that:

http://64.124.13.3/Images/pfssg_4.png

The first spike shows about 90% system overhead when using a 'usleep 100' in the loop? Almost no overhead when running the loop with no sleep. This is a serious flaw in the scheduler?

Would it be worth it to open a new thread pointed to the task scheduler??

MrUmunhum · 11-03-2015, 02:38 PM

Quote:

Originally Posted by MrUmunhum

The images are on my server at 64.123.13.3. Try these links:

http://64.124.13.3/Images/pfssg.png
http://64.124.13.3/Images/pfssg_2.png

The second one is very interesting. It shows the outrages system overhead. More overhead than real work.

I ran this again on a 4 CPU machine with almost no real work being done. Here is the plot for that:

http://64.124.13.3/Images/pfssg_4.png

The first spike shows about 90% system overhead when using a 'usleep 100' in the loop? Almost no overhead when running the loop with no sleep. This is a serious flaw in the scheduler?

Would it be worth it to open a new thread pointed to the task scheduler??

The IP address for my server has changed, it is now:

mt-umunhum-wireless.net. 49 IN A 216.173.131.138

I am reopening this issue since it has not been resolved.