Veeam VMware: Virtual Machine CPU Usage Analysis monitor

[keithkleiman]

I notice that this monitor generates a wanring/critical alert after 1 sample. Can I assume since my collection interval is set to 5 minutes then this would mean that this monitor is collecting a sample every 5 minutes? If so I think it would make sense to probably tune this to 3 samples and change to only alert on critical if I only want to be alerted when CPU is in a critical state for more than 15 minutes.

[Alec King]

Hi Keith,

Certainly - all our monitors support multiple sample values, you can override that in the usual way. If you set NumSamples to three, then the trigger metric will be evaluated over 3 samples (=15 minutes) before an alert is generated.

I've one point to make - all our Analysis Monitors contain additional relevant data in the Description, gathered when the alert fires. Sometimes we get data about the same target object, sometimes we get data about other related objects (for example, for a Host performance alert we might get additional performance data for the VMs on that Host)

In the case of monitor VM CPU Usage Analysis, the Alert Description template is -

Code: Select all

The virtual machine [vm name] running on host [hostname] has a CPU usage issue.

The VM has used  [cpuUsedPct]% CPU. Usage is [cpuUsageMHz] MHz over [cpuCount] virtual cores.

(plus some additional analysis on properties, e.g. CPU Limit etc)

If you set NumSamples to three, then the trigger metric [cpuUsedPct] will be calculated over 3 samples (it will show the average value over three samples, in fact)
However the other data - in this case [cpuUsageMHz] - will show the current value, that is, the value from the most recent sample. This is because we grab this additional data from WMI when the alert fires, and WMI only stores the current sample.

Therefore if you increase NumSamples, you may wonder why cpuUsedPct does not convert exactly to the value shown for cpuUsageMHz. It's because cpuUsedPct is averaged over NumSamples, and cpuUsageMHz is the most recent single sample. Especially with metrics as dynamic and fast-changing as CPU usage, you may see some discrepancy.

Actually I plan to update the Alert Descriptions for our monitors to make this clearer! In this case, I'll be adding the text in italics - 'The VM has used [cpuUsedPct]% over [N] Samples. The current usage is [cpuUsageMHz] MHZ over [cpuCount] virtual cores'. That change will be in our next update.

So, as long as you take note of the above, then there's no issue in tuning any of our monitors to higher sample levels. Just override the NumSamples parameter in the usual way.

Cheers
Alec

[keithkleiman]

How about the "Veeam VMware: Virtual Machine Compute Latency Analysis". I do not see an overridable parameter for the number of samples. Any suggestions on this one? I do however see that I can adjust threshold. Thanks

[Alec King]

Hi Keith,

Only monitors with a single trigger metric can use NumSamples - this is due to internal configuration of the data sources within SCOM. As our Compute Latency Monitor uses two metrics to drive it (cpuLatency and memoryLatency) then Number of Samples is not supported I'm afraid.

You can override NumSamples on CPU Usage, Memory Usage, and Storage Latency monitors for VMs, they all use single metric to trigger.

Appendix C in the MP Operations Guide lists all the performance monitors, you can see the trigger metrics for each. It's here - http://www.veeam.com/veeam_mp_for_vmwar ... 6_0_pg.pdf
Any monitor with a single metric will support NumSamples.

Any questions let me know!
Cheers
Alec