host cpu usage versus physical cpu or vm cpu usage

[perf_guy]

We experienced some issues on Monday and are working on driving into forensic evidence on what may have happened.   We have a Windows host where we see differences in the cpu usage %, peaking about 80% or so on average, but the metrics of both physical cpu (on the host page) or the vm cpu usage % on data explorer are showing the average pegged at 100%.   The reason for the latter two metrics is "physical cpu" is not in data explorer and vm cpu usage % is not a metric on the host page, but I think there are the same.   But my question, was is the difference between these two sets of metrics, the host "cpu usage %" and the, in my case, higher consumption of "physical cpu" or "vm cpu usage %"?

Thanks.

[AntonioSousa]

@perf_guy,

Just to clarify, I believe you are referring to a Windows host A, that has a VM running inside it, that I would call VM B. You have OneAgents installed on host A and VM B. Is this the scenario, Or are we talking about the VMware integration?
In any case, you can post graphs here; if you could show us an image of what you are referring to, it would help.

[perf_guy]

I'm dealing with a Windows VM running as a guest on a VMWare ESXi host.

[AntonioSousa]

Managed to get an excellent example of this one again:

1. CPU usage seems pretty high, but not that high:
   
   
   
2. Physical CPU is well, at 100% most of the time?
   
3. In Classic view, you see the line disappearing, so something is happening:
   
   
   
4. And yes! the new Infrasctructure & Operations app shows this nicely:
5. 1 minute resolution in Data Explorer, with a min/max 0,120% , also confirms it going above 100% most of the time:
   

To be clear, these are values that are being sent to Dynatrace by VMware. Dynatrace does cap the CPU graphs at 100% in classic and in the Unified Analysis view, and so hides this a little bit. Good reason to migrate to the new apps.

I have been trying to find an explanation to this behaviour, it seems to lie around MHz calculations. If I find something will post it back...

[AntonioSousa]

It gets even more interesting when you look at several metrics at the same time, for the same host:

As can be seen, VMware CPU usage and OneAgent CPU usage agree along time. Indeed, a division of the two gives a pretty stable ratio.

Now, the same cannot be said about the VMware CPU usage MHz. While the two metrics above dropped significantly in the end of November, the MHz has remained pretty stable, and didn't drop there. Why? Because the VM was given more cores: amount of CPU used remained stable, but percentage went down 🤣

But, indeed, when we talk about CPU % usage, a lot more has to be told 😁