CPU load - something strange ... help?

[sypher01]

Hi guys,
i'm seeing something strange here.

I'm getting a consistent trigger of cpu load over 5 for a server that has cpu usage at 0-1%.

[root@localhost ~]# zabbix_get -s servername -k system.cpu.load[,avg1]
10.100000


Server is a win2003 SE SP2, i don't understand whats happening.
The reported value is changing, up and down, but i can't find any correlation to the data i see on the actual machine.

[guzzijason]

Hi guys,
i'm seeing something strange here.

I'm getting a consistent trigger of cpu load over 5 for a server that has cpu usage at 0-1%.

[root@localhost ~]# zabbix_get -s servername -k system.cpu.load[,avg1]
10.100000


Server is a win2003 SE SP2, i don't understand whats happening.
The reported value is changing, up and down, but i can't find any correlation to the data i see on the actual machine.

First, cpu "usage" and cpu "load" measure completely different things - don't confuse the two.

Second, how many logical cpus does your system have? The item key you show above is getting the total run queue across *all* cpus, which is misleading. You probably want to use the 'percpu' option so that you can see the normalized load across all cpus, like so:

system.cpu.load[percpu,avg1]


__Jason

[sypher01]

Hi Jason,
one cpu; however as i said, from the values in the perf mon i can see the load but it is different (Albeit high still).

I don't understand what's going on with that machine.

[guzzijason]

Unfortunately, I don't use Windows at all, so I can't comment on what perf mon may or may not be reporting. Hopefully, someone with more Windows experience will chime in here with some pointers.

But as I alluded to before, be sure you're looking at what you think you're looking at. CPU "load" refers to the number of processes in the run queue, and CPU "utilization" (or processor time) is something entirely different. If you are expecting a correlation between the two, don't - because there is no correlation. It is possible to have a low-usage CPU still have a relatively large load (run queue), and likewise, you can have a ~100% utilized CPU still report a low load number.

__Jason