SAP on VMWare: Can you trust the CPU seconds reported by ST03?

[Former Member]

It is a common problem on virtualized systems that you cannot trust the CPU utilization reported from a monitor running within a virtual machine (logical partition).

My question: Can we trust the CPU seconds and response time metrics reported in ST03 when the SAP instance is running in a VMware virtual machine?

Any information or discussion helps. Thanks.

--

Tim

[Former Member]

Beats me, but is this something you can determine by testing?

Rob

[Former Member]

> is this something you can determine by testing?

I'm sure we'll end up doing some testing. I'm interested whether anybody else has looked at the issue.

I know IBM Power5/6 platforms have support for collecting CPU usage. System monitors, like NMON, use those collectors to report physical usage of virtual machines (LPARs.) Application running within the VMs can also access those collectors.

On the Power5/6 platforms, we have compared the CPU seconds reported by ST03 with the utilization reported by NMON and are convinced that SAP is using reporting actual physical usage.

We haven't yet done a similar test for VMware on i86 platforms. I know VMware ESX can report physical CPU usage of the VMs, but I don't think that percolates up to the applications (e.g. SAP) running within the VM.

I know that you cannot trust the process utilization metrics reported by PerfMon within a VM. You can be assured that if one process is using twice the CPU of another, that the ratio is valid. But you cannot trust the actual numbers.

The analogous case for ST03 is that if one step is reported as using 2 sec of CPU and another is using 1 sec of CPU, you can trust that the first is using 2x CPU of the second. But you can trust that actual 1 sec and 2 sec numbers.

--

Tim

[Former Member]

Closed thread until I find more information.

Thanks.

--

Tim