I have been monitoring all our dedicated servers for over a decade with MRTG and the CPU usage is accurately measured with this tool.

Our company is just starting out with the VPS model and we are testing out linode services.

Today we got an alarm via email saying our CPU exceeded 90%. I logged in and the linode dashboard (and longview) showed the CPU pegged at over 240% for over 1 straight hour. But our MRTG CPU graphs showed the CPU at only 1% during that entire time.

I know exactly what the process was and why it shot up. My question for the forum is if anybody has experience with measuring CPU usage on these nodes with MRTG. We are obviously not doing it right and when we opened a support ticket, they told us they did not know MRTG and that we should as for help on the forums.

Any help is appreciated.

The MRTG string we are using is:
Target[m_cpu]: ssCpuRawUser.0&ssCpuRawUser.0:public@localhost + ssCpuRawSystem.0&ssCpuRawSystem.0:public@localhost + ssCpuRawNice.0&ssCpuRawNice.0:public@localhost

Thanks.

As far as I know, nearly everybody migrated away from MRTG years ago (this may be why Linode support is unfamiliar with it); most people around here use munin, since it's pretty much a "set it and forget it" that monitors lots of stuff. Example:

http://gus.fixppp.org/munin/localdomain ... index.html

I believe the heritage of Munin is that there was MRTG, and then there was RRDTool which was a re-implementation of MRTG to address its many limitations, and then Munin was forked from that.

Fun fact: My file server and home theatre projector share the same UPS, so you can tell from the graphs when I'm watching stuff on my home theatre system

I never could get the hang of SNMP strings. Instead, I use an external script with MRTG (for over a decade, wow).

Script is at http://parsed.net/mrtg/mrtg-data
Results at http://parsed.net/mrtg/