Gaps/spikes in CPU graph, other graphs ok.

[hoganw]

So, I noticed the other day on a newer server that 'top' on SMP servers has the ability to display either average or cumulative CPU usage...taking that once step further, I thought net-snmp might handle CPU usage the same way.

Ta-da!

On a 4-cpu Xeon HT system, Linux thinks there are 8 CPUs so it reports up to 800% CPU usage via net-snmp. For these machines I created a new template with a CDEF to divide by 8, presto! Graphs that report 0-100% CPU. The other hitch is that the template also has to have the max increased from 100 to 800, which was the "real" fix to the gaps, the CDEF just gets the graphs to display 0-100.

On a 2-CPU Xeon HT it sees 4 CPUs, so reports up to 400%, CDEF that divides by 4, and so on.

Anyone else running SMP servers you need to fix your templates. Even on graphs that don't LOOK broken they really ARE broken due to how net-snmp reports CPU usage. A dual CPU system that shows CPU graphs with a 20% average, the real number is 5%.

[macindy]

Sorry can you explain again how you fixed the gaps?

1.) Set the limit to 400 in the graph template (for a DualXeon))
2.) Set the limit to 400 in the data templates (nice, user, system)?