Managing guest memory - Virtual machines | OpenShift Virtualization

Configuring guest memory overcommitment
Disabling guest memory overhead accounting

If you want to adjust guest memory settings to suit a specific use case, you can do so by editing the guest’s YAML configuration file. OpenShift Virtualization allows you to configure guest memory overcommitment and disable guest memory overhead accounting.

The following procedures increase the chance that virtual machine processes will be killed due to memory pressure. Proceed only if you understand the risks.

Configuring guest memory overcommitment

If your virtual workload requires more memory than available, you can use memory overcommitment to allocate all or most of the host’s memory to your virtual machine instances (VMIs). Enabling memory overcommitment means that you can maximize resources that are normally reserved for the host.

For example, if the host has 32 GB RAM, you can use memory overcommitment to fit 8 virtual machines (VMs) with 4 GB RAM each. This allocation works under the assumption that the virtual machines will not use all of their memory at the same time.

Memory overcommitment increases the potential for virtual machine processes to be killed due to memory pressure (OOM killed).

The potential for a VM to be OOM killed varies based on your specific configuration, node memory, available swap space, virtual machine memory consumption, the use of kernel same-page merging (KSM), and other factors.

Procedure

To explicitly tell the virtual machine instance that it has more memory available than was requested from the cluster, edit the virtual machine configuration file and set spec.domain.memory.guest to a higher value than spec.domain.resources.requests.memory. This process is called memory overcommitment.

In this example, 1024M is requested from the cluster, but the virtual machine instance is told that it has 2048M available. As long as there is enough free memory available on the node, the virtual machine instance will consume up to 2048M.
```
kind: VirtualMachine
spec:
  template:
    domain:
    resources:
        requests:
          memory: 1024M
    memory:
        guest: 2048M
```
The same eviction rules as those for pods apply to the virtual machine instance if the node is under memory pressure.
Create the virtual machine:
```
$ oc create -f <file_name>.yaml
```

Disabling guest memory overhead accounting

A small amount of memory is requested by each virtual machine instance in addition to the amount that you request. This additional memory is used for the infrastructure that wraps each VirtualMachineInstance process.

Though it is not usually advisable, it is possible to increase the virtual machine instance density on the node by disabling guest memory overhead accounting.

Disabling guest memory overhead accounting increases the potential for virtual machine processes to be killed due to memory pressure (OOM killed).

Procedure

To disable guest memory overhead accounting, edit the YAML configuration file and set the overcommitGuestOverhead value to true. This parameter is disabled by default.
```
kind: VirtualMachine
spec:
  template:
    domain:
    resources:
        overcommitGuestOverhead: true
        requests:
          memory: 1024M
```
If overcommitGuestOverhead is enabled, it adds the guest overhead to memory limits, if present.
Create the virtual machine:
```
$ oc create -f <file_name>.yaml
```