OpenVZ Forum


Home » General » Support » Monitoring VE's - Load is important?
Monitoring VE's - Load is important? [message #50882] Tue, 19 November 2013 22:18 Go to next message
grep is currently offline  grep
Messages: 34
Registered: November 2013
Member
Hello,

i have a really simple question.

I written a bash script which get from all nodes vzlist -o veid,laverage and report if a VE has too much load.

But: When i already limit the CPU usage of the VE's to 10% i think the load is not meaningful about overloading the node, correct? It just say that the CPU which is dedicated to the ve is overloaded but not the node?

If the VE has a load of 10, get the load of the node increased by the node because there is a large cpu queue of one ve or does this not increase the node load?

And i think if the load on the node get higher trough the high load on the node it not must say automatically that the node is overloaded. The important value is the CPU usage of the node, is this correct?

So if load on node is 20 because one VE is overloaded and the used cpu is only 15%, then nothing is overloaded, right?

Is there a option to get a value how much % of cpu the VE takes from the node? in user_beancounters is no such value.

I am happy about a detailed answer. Thanks a lot!
Re: Monitoring VE's - Load is important? [message #50908 is a reply to message #50882] Wed, 27 November 2013 22:37 Go to previous messageGo to next message
grep is currently offline  grep
Messages: 34
Registered: November 2013
Member
push, sorry Smile
Re: Monitoring VE's - Load is important? [message #50909 is a reply to message #50882] Thu, 28 November 2013 06:50 Go to previous message
xaxaxa is currently offline  xaxaxa
Messages: 11
Registered: August 2012
Location: Vancouver, BC
Junior Member
from my experience load avg means nothing; if i run a fork bomb in a container, the load average on the host quickly shoots up to thousands, even if i've only assigned one cpu core to the container (and cpu usage on the host is 25% on a quad core system)

personally i think the best way to detect "overload" is by "pinging" a service you're running, and measure latency (for example send an http request)
Previous Topic: dcachesize possible issue?
Next Topic: mainline kernel
Goto Forum:
  


Current Time: Sat Jun 22 02:26:05 GMT 2024

Total time taken to generate the page: 0.02926 seconds