CPU Ready time

[Former Member]

Hi everyone

we making load test on our enviroment (many application server, one db-server, all on vmware esx 5.5) and Monitoring, inter alia, the cpu ready time on the db-Server vitual machine. the Tools to Monitor are the wily introscope (hostagent/sap oscol on the guest host) and esxtop/vCenter on the vmware-side.

Now we have different results:

on wily, we see the timeframe of tests with Peaks from 400 to 600 % (ok --> value is *10)

On a snapshot with esxtop at 14:58 (highst load on wily), we have a readytime on the same vm of 0.78%

And on vCenter, we dont see any heavy load on the vm during the load test.

Which values are correct? Is hostagent/sap oscol meassurement are correct? i have the guess, that the hostagent collect wrong data.

Any explanations about the behaviour?

Regards

Thomas

[former_member226644 · Accepted solution]

Hello Thomas,

we have investigated this behavior. In the VMware tools Guest SDK 5.5, there is only the "Steal Time" metric. This metric is used to generate an approximate ready time. This number should be comparable to "ready time" seen in vCenter and on the ESXi host.

In the investigation, we found out, that this is not true for all occasions, since "ready time" includes some metrics, which are not found in "steal time".

That means, that there can be a difference between these two numbers and in doubt, always use the one seen in vCenter or on the ESXi host.

In the new releases of ESXi (5.5u3 and 6.0) a new metric was introduced, "contention time" this metric is more accurate and will also be enhanced if needed, like when new datapoint have to be used, to increase the accuracy of it.

So, as for now, please use the SHA/SAPOSCOL numbers as an indicator for you. If this number goes up, check the "ready time" in the vCenter for the final conclusion.

Best regards

Sebastian Lenz

VMware

[former_member307673]

Hi Thomas,

Many thanks for sharing this analysis! Our tools - in this case the SAP Host Agent (SHA) with included SAPOSCOL that "feeds" the Wily IS - rely on the data provided by API delivered with VMware tools.

So:

1. I believe that you can rely on esxtop/vCenter/vcops-vrops as these are "vendor tools". For future please add the vcenter statistics to the set of screenshots.

2. Of course it might be that SHA/SAPOSCOL has an issue - e.g. maybe a kind of overflow. First recommendation would be that you use latest SHA version (can be updated anytime) and using latest version of vmware tools for the corresponding VM.

Even if the update (2.) is not possible we can support you in doing a deeper analysis. For this it would be perfect that the issue is reproducable and you create a SAP incident (refer to to this thread and my name). Then we will debug and deeply analyze SHA/SAPOCOL. When doing so please create a document that contains the analysis on each level as you did with the screenshots.

BR,

Sebastian

BTW: Of course we expect that the values on each level are comparable as Ready Time is a very important metric for us.