CPU Ready per CPU not per VM

[stingray75]

Hi,

Would it be possible for Veeam to report/alarm on CPU Ready per CPU, rather than per VM? I ask because i'm getting alarms that CPU ready is over 10% for certain VMs. However, the VMs in question have 4-6 vCPUs, so average CPU ready per CPU is between 1.6 and 2.5% which isn't a problem at all. If we had a new counter, say CPU Contention %, which takes into account CPU Ready, number of vCPUs etc, it would be far more useful.

Thanks

Matt

[Vitaliy S.]

Hi Matthew,

Hmm...I think it would be possible to count the number of vCPUs and calculate the average CPU Ready for the VM. Let me put it to the feature tracking system.

Thanks for the feedback!

[thomasl]

Hi Vitaly,

just to make sure if I understand this posting right in relation to our installation (ESXi 5.0, Veeam ONE Monitor 7.0) and some observations we made:

Right now we are in a process of migrating some bigger workloads from Physical to VM, each of them with 8 and more vCPUs. We don't have overcommitment of ressources, there are enough pCPUs available.
We are getting more and more "VM CPU ready" alerts especially from these large VMs, telling us that there is sometimes more than 20%/30% "CPU Ready" in that VM for a 15 minute time frame. At the same time no end user is lacking performance and even with other monitoring tools (inside the hosted applications) we can't find serious problems with the performance.
When I assume that Veeam Monitor is summarizing the percentage of "CPU Ready" for each vCPU - we really don't have a problem: a VM with 16 vCPUs with 1-2% CPU ready for each vCPU will be reported at 25% CPU ready and Veeam will fire an alert. Is that the way Veeam is working right now?

I only want to be sure if my conclusions are right (we have to re-define the alerts then).
For the future it would be nice if Veeam Monitor would consider this during its calculations like stingray75 already said.

Thanks
Thomas

[Vitaliy S.]

Hi Thomas,

Yes, your assumption is right. CPU Ready is tracked per instance of the VM and not by the vCPU. I have asked to change this behavior in one of the next hotfixes.

Thanks!

[Vitaliy S.]

Just an update - our support team has a private hotfix that would enable per core CPU monitoring thus all these alarms will be triggered based on the core value, and not based on the sum of all cores performance.

If you need this fix, please contact our support team directly, as this will NOT be included to the upcoming patch/Veeam ONE v7 R2 update.

[init-s]

Hi to all,

I'd like to understand better the situation about CPU Ready and Veeam One monitoring.
Many website wrote that for VMs the best is to have <5% CPU Ready Time and that >10% CPU Ready Time is a problem and must be investigated.

But, as reported by the other user, any vCPU has a CPU Ready Time.

So where is the truth?
A VM must have a CPU Ready sum or a CPU Ready per vCPU < 5%?

Thanks

[Vitaliy S.]

If you're using a single vCPU then the logic for entire VM can be applied. However, if you have many vCPUs for the VM and it is extensively using all of the vCPUs, then you should be monitoring CPU Ready per each vCPU. Also it depends on how you calculate CPU Ready for entire VM, if it is going to be the sum of all CPU Ready values for each vCPU, then you will definitely go above 5% very quickly.

[albertwt]

So isthere any way to display this report now in VeeamONE v8.0 ?

[Vitaliy S.]

Albert, can you please clarify which report you want to see? In v8 both alarms and existing reports are based on CPU Ready value for the entire VM, in previous version we were using per core CPU Ready calculation which led to false alarm triggers.

[albertwt]

Albert, can you please clarify which report you want to see? In v8 both alarms and existing reports are based on CPU Ready value for the entire VM, in previous version we were using per core CPU Ready calculation which led to false alarm triggers.

Ok so I this case with Veeam ONE v8.0 the VM CPU ready is already divided per vCPU core in percentage ?

[Vitaliy S.]

Yes, in v8 CPU Ready is an average value across all cores of the VM.