VM vCPU time per dispatch on two virtual machines

[burak]

Hello,

I recently installed Veeam ONE to monitor backup server and two Hyper-V Hosts. Now, I get periodically alarms on two VMs that the VM vCPU time per dispatch is over the warning threshold. First, the time was mostly a little bit over 60ms, so I reconfigured the warning threshold to 65ms. Now, the value is between 65 and 80. Today, I got 20 alarms for one VM and 10 for the other. Is this something I should take care of or is it negligible?

The Server 2012 Hyper-V host has 2x Intel Xeon E5607 @ 2,27GHz with 4 cores each and NO hyperthreading. This means 8 cores together. I have 6 VMs with Windows Server 2008 R2, where 5 of 6 has 2 vCPUs configured and one VM has 1 vCPU. Both VMs that report the dispatch warning has 2 vCPUs. Yesterday, I switched them both from 4 vCPUs to 2 vCPUs but the warning is still reported.

The other VMs hasn't reported anything until I installed Veeam ONE (its been 2 or 3 weeks). The both VMs that report this warning services as file server and WSUS server.

Any recommendations what I should look after? Is the vCPU configuration bad? I think giving a VM one vCPU only is a little bit less, it's like running Windows on an old one-processor one-core system. At least if Windows Updates are running, the system will go really slow.

Thanks!

Best regards
Burak

[Shestakov]

Hello Burak, CPU time per dispatch tells you how much time virtual CPU is waiting for command execution from logical(physical) CPU. It`s not critical, but slows down overall performance.
First thing to check is vCPU/lCPU ratio which should be less than 8. I would recommend to run Hyper-V Configuration Assessment report and VM optimization reports to find VMs with excessive number of vCPUs.
Thanks!

[burak]

Hello,

thanks for your answer. I run the Hyper-V Configuration Assessment report and got for Virtual CPUs/Logical CPUs ratio a value of 0.63. I did not find VM optimization report but Overseized VMs report give me the info that four virtual machines (which includes the both file server which generate the warnings) should be configured with 1 vCPU instead of 2 vCPU. I really can't agree to this recommendation, because using one vCPU on a virtual machine is , like I said, really less, especially when the server is performing multiple operations. Also, using 11 vCPU on 8 lCPU is in my opinion not a big overcommitment.

Maybe the processors are not that fast and the warning of 60ms is more for processors of newer generation? Intel Xeon E5607 is from 2011. Should I maybe just ignore the warnings and set the warning and error threshold higher?

Thanks!

[Shestakov]

0.63 is not a critical value at all, so vCPU/lCPU ratio is not an issue in your case.
Another thing affecting the dispatch time is non-virtualization related applications running on the host.
You are right that hardware productivity matters, I would rather increased the thresholds instead of disabling the whole alarm.