Dangerous CPU Utilization Uptrend

Discussion of the Meteohub software package

Moderator: Mattk

Post Reply
Headworx
Senior Boarder
Senior Boarder
Posts: 45
Joined: Thu Sep 17, 2009 10:08 pm

Dangerous CPU Utilization Uptrend

Post by Headworx »

Hi, I run MeteoHub on a SheevaPlug with 4GB 30Mbps Sandisk SD card. There are three weather stations defined (two RFX receiver and a Davis) plus virtual "system data" station.

I experience steady rise of CPU utilization on the system, now approaching 60%... What can be the reason and what can be done to stop this trend?

Attached is a relevant graph.

Image

Cheers,
Headworx
User avatar
d_l
Expert Boarder
Expert Boarder
Posts: 140
Joined: Fri Mar 13, 2009 12:58 am
Location: Reno, Nevada USA
Contact:

Re:Dangerous CPU Utilization Uptrend

Post by d_l »

It's standard for the system load to steadily rise through the month until the end, but then it resets to a lower level in the first several days of the following month. This is on the NSLU2, but I guess it must apply to all platforms.
Vetinari
Junior Boarder
Junior Boarder
Posts: 39
Joined: Tue Aug 11, 2009 9:01 am

Re:Dangerous CPU Utilization Uptrend

Post by Vetinari »

I had it as well. Make a reboot, perhaps this will help. It worked for my meteoplug.

But btw. how can you use the system settings to be displayed? Add the system-station under the weatherstation-side?
Headworx
Senior Boarder
Senior Boarder
Posts: 45
Joined: Thu Sep 17, 2009 10:08 pm

Re:Dangerous CPU Utilization Uptrend

Post by Headworx »

Looking at the 3-month chart, I see b_l seems to be right... the load rises through the month and then falls down when the next month begins.

To graph system data, just add a system data virtual station and start collecting data points. The important ones for me are the system load and storage.

I am currently working (time permitted) on a simple plugin to monitor other hosts on the network and use MeteoHub as a generic data collection and graphing engine, plus the alerts to send an email when something goes wrong...

Image

Cheers,
Headworx
User avatar
d_l
Expert Boarder
Expert Boarder
Posts: 140
Joined: Fri Mar 13, 2009 12:58 am
Location: Reno, Nevada USA
Contact:

Re:Dangerous CPU Utilization Uptrend

Post by d_l »

Headworx, you've probably been adding chores for your meteohub to porocess over the past three months so your peak load at the third month is much higher than the peak of the first month.

Another thing that seems to excessively increase my system load is having bad data hanging around in prior months. I periodically use the Cleanup Data button under the Inspect Data section on the current month's data. I especially do this if I've had a spontaneous reboot.
Headworx
Senior Boarder
Senior Boarder
Posts: 45
Joined: Thu Sep 17, 2009 10:08 pm

Re:Dangerous CPU Utilization Uptrend

Post by Headworx »

I have 3 weather stations (1 Davis and 2 RFXComs), 6 sensors each (that totals to 18 real sensors + 6 additional for system data). The performance of the system (web interface, graphs...) is very good, I have just been wondering if it will settle somewhere around 50% or will continue to climb through the roof... will see in 3 days, when November ends... :)

Have not used the cleanup data... but have not had any reboots either... Sheeva is rock steady... I'll wait till December and then may be will do the cleanup, depending on the load.

Thanks for yout tips :)

Cheers,
Headworx
User avatar
HeinrichH
Gold Boarder
Gold Boarder
Posts: 187
Joined: Sun Jan 04, 2009 6:27 pm
Location: Emmen-Netherlands / Panoias-Portugal
Contact:

Re:Dangerous CPU Utilization Uptrend

Post by HeinrichH »

Maybe a stupid question but how do you read those data fields?
When I'm looking at the system info tab I see a system load of 0,54 0,56 0,57 but when I'm looking at the sensor page the processor load is 0,58.
What do those numbers mean?

How did you defined your graph? What's the time bucket and what are the units used?

Thanks in advance, Henk
Checking the remote....
Headworx
Senior Boarder
Senior Boarder
Posts: 45
Joined: Thu Sep 17, 2009 10:08 pm

Re:Dangerous CPU Utilization Uptrend

Post by Headworx »

0,54 is 54% load...

My graph is defined as a scalar data on a time line, for the last 100 days, with 6 hour buckets. I use data0 sensor for CPU load. Beforehand there is a Weather Station 1 of System Data type defined, that provides the necessary sensors. Pretty standard stuff... I think the graph and values are correct. Today I have 60%, lets see what comes up tomorrow...

Cheers,
Headworx Image
Headworx
Senior Boarder
Senior Boarder
Posts: 45
Joined: Thu Sep 17, 2009 10:08 pm

Re:Dangerous CPU Utilization Uptrend

Post by Headworx »

It has dropped now to 10%... So it really looks like it is a monthly cycle...

Cheers,
Headworx
User avatar
d_l
Expert Boarder
Expert Boarder
Posts: 140
Joined: Fri Mar 13, 2009 12:58 am
Location: Reno, Nevada USA
Contact:

Re:Dangerous CPU Utilization Uptrend

Post by d_l »

For some reason on my NSLU2, the lowest load levels are actually reached on the second day of the new month, not the first. Do you see the same?
User avatar
HeinrichH
Gold Boarder
Gold Boarder
Posts: 187
Joined: Sun Jan 04, 2009 6:27 pm
Location: Emmen-Netherlands / Panoias-Portugal
Contact:

Re:Dangerous CPU Utilization Uptrend

Post by HeinrichH »

I don't know, started with system logging only 48 hours ago, I will inform you over a day of 40 or so :-)

Henk
Checking the remote....
Headworx
Senior Boarder
Senior Boarder
Posts: 45
Joined: Thu Sep 17, 2009 10:08 pm

Re:Dangerous CPU Utilization Uptrend

Post by Headworx »

Yep. This is the case.

On Nov 30 it was 60%.
On Dec 01 it was 18%.
On Dec 02 it was 10%
Today (Dec 3) is 12%

On the other hand my month highs were:
40% in September
50% in October
60% in November

Wonder if the end of month maximu CPU load has to do with the fact I kept on adding sensors and weather station to the system or because the amount of historical data was rising?

Anyway... will see by the end of December... As I have not added any new sensors / weather stations during recent weeks...

Cheers,
Headworx
Post Reply