Inconsistency in Cacti vs. sar/vmstat/top measurements ?

[Ingimar]

Hi there.

We're measuring the CPU of a dual Xeon machine and are getting what we think are wrong (or at least not right ) numbers.

It seems that the Cacti measurements are much too high compared to sar (from the sysstat package) measurements. Also when top and vmstat snapshots are compared they seem to be quite consistent with the sar output.

Could it be that being dual Xeon machine that net-snmp is measuring 4 "ticks" for each "tick" and thus the Cacti graph showing 4x the correct numbers?

Attached are a graph from Cacti and sar output from midnight to ~ 14:00 today.

[gandalf]

There has been a lot of rumours about the SNMP CPU measuring. Please see past forum traffic on this
Reinhard

[ehall]

On my Xeon Linux system, I frequently see processor times well over 100%. It's because the two logical CPUs are being added together.

If you read the net-SNMP FAQ you'll see that support for multiple processors is fairly weak.

This is largely due to weaknesses in the underlying OS. If you're using Linux, look at some of the files in /proc and you'll see what I mean.

This is all rather unfortunate, but unsurprising, given the near-total absence of SNMP for *NIX applications and services in general.