vCPU vs GHz

[ASHISHB_SHARMA]

ESXi host that I have has the following specs

2 sockets

16 cores per Socket@ 2.10 GHz

Logical Processors are 64

Total Capacity is 67.04 GHz

I want to understand the total number of vCPU which can be allocated to VMs on this host. Moreover what is the relation between the available CPU with the number of active VMs on the host. And what will happen in case if I allocate total number of vCPU to VMs which is greater than the available number of logical processors i.e. 64 in this case.

[sjesse]

Look at https://configmax.vmware.com/  and change the filter to vsphere and esxi host limits, these the only limits that matter. You can have more vcpus that the logical ones that are available, but as you would expect the their share of the resources get smaller, If you do research on line you'll see people talking about a vcpu to pcpu ration, in some cases people have 10 virtual cpus per phyisca ore and that works fine, but in other cases people have 2 virtual cpus per phyiscal core.

pCPU vs vCPU - Relation/Ratio

[ASHISHB_SHARMA]

Thanks sjesse for your response.

I have got information from the link that you have provided. Going through the link that you have provided https://configmax.vmware.com/  I see that vSphere 6.7 allows for 256 Virtual CPUs per virtual machine & Virtual CPUs per host maximum is 4096.

In this instance what I am interested in is the limit which can be defined based on the max limits. To give a better idea I have 26 VMs on the host at the moment with 75 vCPU allocated to them whereas there is only a total of 64 Logical processors that are available. Hence my question around the impact that it can have on the performance of the VMs. Although the available capacity in terms of CPU is 60.07 GHz out of the total of 67.04 which is available in total.

[sjesse]

Cpu are given time in shares, if you 64 Logical processors that are available and you had 64 vcpus then each vcpu would be able to work on what they need 100% of the time, though they only get 50% of the processing power of the core. Anything above that the cpus time get scheduled , this is where the cpu ready metric comes in.Cpu ready time basically means the vcpu wanted to do something but it couldn't so it needed to wait, If you only use the 64 vpus on your hosts you should basically have 0 cpu ready time. You shouldn't worry until you get around a 5% ready time per vcpu. Some apps are more sensitive to cpu latency so it important to test, as no one number works for everyone. Take a look at ESXTOP – Yellow Bricks | Yellow Bricks if you haven't as these are some general guidelines for the metrics that esxtop provides

[scott28tt]

Chapters 2-4 of this are worth reading: https://docs.vmware.com/en/VMware-vSphere/6.7/vsphere-esxi-vcenter-server-67-resource-management-gui...