Simple problem: snmp cpu-utilization

[lukas]

Hi all!

Is anybody reading the mailinglist? Hey, come'on, mailinglists are the best way to do all this sort of things! ;)

Anyway, my problem is that I monitor the cpu-utilization of my boxes via snmp.
But they all show up all the time with exactly the same percentage:
one server is constantly at 8% User, 5% System, anotherone is at 0% User, 2% System.

I'm using Cacti 0.6.8a on Debian/Woody, net-snmpd on all boxes and am polling OID's enterprises.ucdavis.systemStats.ssCpuUser.0 and enterprises.ucdavis.systemStats.ssCpuSystem.0.

Any comments?
Thanks :)

[bulek]

I prefer this board to mailing list because like you said it is "the best way to do all this sort of things"

Seems you have very stable linux environment (even too stable). How does the utilization look like when you check it with snmpget directly?

- bulek

[lukas]

Hm. Maybe we should write a script to tie the mailinglist and the board together ;)

Well, the environment is not so stable, Load wanks between 0.1 and 2, total no of processes between 80 and 200. When I do a top, then the cpu percentage differs, too. But when I ask snmp, it's always the same.

A strange thing I notice: the initial value of a top is always the same value as snmp says, and after the first refresh the values start to grow or shrink. Hm.

[raX]

I would recommend using the following ucd-snmp OIDs for gathering CPU utilization:

Code: Select all

enterprises.ucdavis.systemStats.ssCpuRawUser
enterprises.ucdavis.systemStats.ssCpuRawNice
enterprises.ucdavis.systemStats.ssCpuRawSystem
enterprises.ucdavis.systemStats.ssCpuRawIdle

These must be added as a COUNTER instead of a GUAGE however. I have found that this method is much more accurate and even reflects CPU usage on multiple processors.

-Ian