High CPU Wait time

[andvm]

Hi,

Is there any technical reason why a VM running a virtual Firewall with plenty of available resources would show more than 1% or 2% of CPU Wait time? (Host has only this VM running and more physical cores that the VM is assigned).

My understanding in this case should be that whenever the VM requests CPU time it should always be served immediately as no other VM is competing.

[Finikiez]

How do you measure 1-2%? esxtop , performance graphs or what?

Can you show a screenshot?

As well can you describe your configuration?

[andvm]

esxtop, would there be any reason for this that can be explained or referenced please?

[Alex_Romeo]

Hi,

You should provide some more information.

What Firewall does it perform?

What Esxi version?

It is not always true that more vCPU means more speed.

I agree with the other answers on esxtop:

https://docs.vmware.com/en/VMware-vSphere/6.7/vsphere-esxi-vcenter-server-67-monitoring-performance-...

Best regards,

Alessandro Romeo

[andvm]

Opensense on 6.5U2

[Kev_Johnson]

What is your host hardware, and do you have any other VMs running on that host?

[andvm]

at least 32 cores (VM has less than 10vcpu assigned so this is the part which I cannot understand), vmotioned VM to other hosts and VM CPU wait time does not change, gets worst of course if there is contention from other VM's.

The 2% I am referring to is when there are no other VM's running on the same host, expect this to be 0.x and not this high.

Any ideas why and confirm if what I expect is correct?

[sjesse]

How may cpus does the vm have, all vcpus from a vm wait to be scheduled at the same time, so if the vm is only doing work on one vcpu and 5 of them are sitting there you can see higher wait times. I've seen improvements by lowering the cpu count in a lot of cases.

[Finikiez]

So first of all WAIT includes IDLE (the time when VM does nothing)

As well in esxtop you see summary of all worlds by default. If you expand GID then you'll see counters by each vcpu and other world. I expect that the numbers will be 0.xx as you expect.

See Why is %WAIT so high in esxtop?

[andvm]

Apologies for the confusion, I  really meant %RDY  when seen via esxtop (Which itself is the time the VM was waiting for available Host CPU resources)

So to me it is not clear why a VM of let's say 10vCPU residing on its own on a host with 32 physical cores would have a %RDY time of 2% or more.

[Finikiez]

The same with %RDY: expand VM by it's GID and see %RDY for each vcpu and other world, they should me small.