vCPU calculation

[Kim_Pedersen]

Hi,

We are running vRops 6.6 and we have a problem with calculation of vCPU's within some specific clusters. Everything was working fine until we saw this:

If I look at the time the problem occur, it's exactly the time vRops is running it's calculation. It's from a Sunday which is not a working day and therefore there has been no changes to the environment - no extra host and if any a very limited number of VM's added. I assume that the number specified in CPU|Allocation|Recommended size (vCPUs) is the total number of vCPUs allocated to the VM's within that cluster ??

We have several clusters and this only occurs on clusters which are in a one custom group there is no problems with the other custom groups. I have seen similar problems in this community and they were not answered but mentioned that this could be because of policy configuration. We have configured policy only for allocation and not for demand.

The problem here is that overnight the value almost doubled and the report to management is all red due to missing vCPUs and I'm not able to give any good reason for this.  

Any ideas would be highly appreciated 🙂

Regards,

Kim.

[sxnxr]

I could be complete wrong about this and have no facts to back it up by my interpretation is

This is a recommended size based on the utilisation of the VMs in the cluster. Even tho you running the allocation based model vrops will monitor the demand of each VM anyway to enable the over/under sizes vm report.

The recommended size will move up and down based on the vms utilisation. As yours is jumping up i would guess that at the time there was high VM utilisation on this cluster causing vrops to calculate that the VMs were running over utilised.

You could try running a couple of reports on the VMs recommended size to see if the data matches up. I dont use the recommended in my reports because of this. I just use the remaining value

[Kim_Pedersen]

Hi sxnxr,

Thanks for answering

If I look at the remaining value as you suggested (I did that as well) It looks like this

If I compare the 2 graphs I can see that amount of free vCPUs is dropping to 0 where the problem occur.

I have tried running a report VM rightsizing and performed a bit of Excel gymnastics and I could see that the total number of recommended vCPUs is approx. equal to the amount before we had the problem on the 11 December. We have even added extra hosts since the 11 December.

Kim.

[sxnxr]

The remaining value will take the below into consideration ( if configured)

1. Any new deployments
2. HA buffers and HA settings. By default the buffers are 10% but can also will take the cluster's HA settings into consideration ( host dedicated for HA as an example)
3. Peak. This will use stress and peaks
4. Committed projects

Changing any of these could impact what you are seeing.

[Kim_Pedersen]

Hi

We have not performed any deployments.

We are not using HA or Peak considerations - we just specify a buffer size.

Since we want to use the allocation - I just wondered if anybody has been able to use metrics for total capacity and used capacity by the VM of both vCPUs and Memory ???

For vCPU we use CPU|Allocation|Total Capacity (vCPUs) and to find the number of vCPUs used we use Summary|Number of vCPUS on Powered On VMs.

We have tried using Memory|Usable Memory and Memory|Allocation|Usable Capacity which should give us the total amount of memory and subtract the buffers. Now we just need to find the total amount used by the VM's and the closest to the correct number is Memory|Allocation|Computed Demand but it doesn't quite add up.

Rgds,

Kim

[Oulyanov]

If your intend is to use allocation with buffers I have found it's more easily to create Super metrics to get the required values calculated

Regards, Oleg

[Kim_Pedersen]

Hi Oleg,

Do you have an example of one of your calculations ???

Thanks in advance.

Rgds,

Kim