We have been experiencing a discrepancy between the Linode manager's CPU graph and our Zabbix CPU graph for a CentOS 6 web server on a Linode 4096.

The linode graph reports around 50% utilisation at our peak web load, and a brief spike of 240% when I ran some dirty benchmarks. Our graphs show 28% at the same time as the linode manager shows 50%, and 380% when the linode manager shows 240%.

I've gone through the Zabbix source code and determined that it uses /proc/stat for analysing CPU utilisation. I performed the same calculations it does in a quick script and found that the numbers I was getting were consistent with the numbers filtering through Zabbix.

The Linode Support team informed me the Linode Manager graphs are "5 minute averages based on 4 cores, so [we] may be running into discrepancies caused by averages versus discrete values, or scaling depending on how Zabbix handles multiple cores."

I'm waiting on some per-core data from Zabbix, which probably won't start filtering through for a few more hours. Beyond that I'm pretty well stumped.

Does anyone have any ideas? Cheers.

Their explanation sounds about right.

_________________
Kevin a.k.a. Dweeber