ESXi CPU resources distribution

cpu-usagevmware-esxi

I have ESXi hypervisor with i7 990x cpu and i want to understand how to provide resources to 4 Virrual Machines with maximum benefit.

First my idea was to provide 1 core for less cpu-dependent vms, and 2 cores for more cpu-dependent vms, and then i thought that while some vms will not use their cores, other vms will run out of cpu resources. Also, i don't know how ESXi provides cores – perhaps it can provide first cpu many times, and last one will never be used.

Is it a good idea to share one cpu between vms? I mean

1vm: 2cores
2vm: 2cores
3vm: 2cores
4vm: 6cores

In sum this is 12 cores, when my processor has only 6. So, i suppose that esxi will share some cores between VMs. Is it a normal practice? Does it have any pitfalls?

Best Answer

Its not a 1:1 ratio for vCPU to pCPU. VMware doesn't map the vCPU cores to physical cores. There is a layer between the VM's and the physical called the scheduler. As the VM's request threads from VMware, the VMware scheduler will assign tasks to the hardware as needed. The scheduler uses VM shares as an indicator of priority. Issuing too many vCPU's to a machine could cause issues because the scheduler is waiting for enough pCPU's to become ready before it can schedule the tasks. If the physical machine becomes overwhelmed, you will see a spike it "CPU Ready".

As a rule of thumb, I always (in most cases) start with 1 vCPU on my VM's. Watch the VMware (not Windows) processor graphs to see if its maxing out. If so, issue another until its average is about 50%.

Related Solutions

Ssh – Mysterious per-session CPU limiting

The Linux scheduler will (if the application doesn't do it on it's own) never bind anything to any specific core. Quite the contrary actually, most of the time the scheduler will bounce the load across all cores at a seemingly random fashion. It is the only part of the kernel which has a kind of holistic view of the system and it tries to "optimize" system response/throughput according some heuristics which in term manifests itself as the kernel bouncing the processes across the cores.

You have to explicitly pin processes to certain cores via something like taskset(1) to accomplish that. Can you provide some evidence for what you are experiencing?

Windows – perfmon – single thread utilizing max CPU – how to detect

Use the WMI processor performance counters. Exclude the counters with _Total, because those average the load.

Get-WMIObject -Class Win32_PerfFormattedData_Counters_ProcessorInformation |
  Where-Object {$_.Name -notmatch "total"} |
    Select-Object -Property Name,PercentIdleTime,PercentProcessorTime

You'll see something like this:

Name PercentIdleTime PercentProcessorTime
---- --------------- --------------------
0,0              100                    0
0,1               95                    4
0,2              100                    0
0,3               89                   10

I'll leave generating the alert up to you.

Best Answer

Related Solutions

Ssh – Mysterious per-session CPU limiting

Windows – perfmon – single thread utilizing max CPU – how to detect

Related Topic