1. Documentation
×

Performance optimization

You can modify certain configurations for your OKD Virtualization deployment to improve efficiency, cost-effectiveness, and reliability.

Configuring full physical cores for virtual machines

As a cluster administrator, you can allocate a full physical core to a specific virtual machine (VM), instead of allowing different VMs to share the same physical core. Configuring your VMs to use only full physical cores can optimize performance for high-throughput or latency-critical VMs.

Allocating only full physical cores is important on simultaneous multi-threading (SMT) enabled systems because it offers the following benefits:

  • Prevents noisy neighbors and resource contention

  • Mitigates performance degradation

  • Offers predictable latency

  • Guarantees exclusive CPU resources

You can configure full physical core allocation by modifying the cpuManagerPolicy and cpuManagerPolicyOptions settings in the KubeletConfig custom resource (CR).

Prerequisites

  • You have cluster administrator access to a OKD cluster with OKD Virtualization installed.

  • You have installed the OpenShift CLI (oc).

  • You have enabled CPU Manager on the node where your VM runs.

Procedure

  1. Edit the KubeletConfig CR to add the required cpuManagerPolicy and cpuManagerPolicyOptions configurations:

    apiVersion: machineconfiguration.openshift.io/v1
kind: KubeletConfig
# ...
cpuManagerPolicy: static
cpuManagerPolicyOptions:
  full-pcpus-only: true
kubeReserved:
  cpu: "1"
# ...

    • You must set the cpuManagerPolicy: static policy to enable exclusive CPU allocation. This setting is a prerequisite for configuring the cpuManagerPolicyOptions settings.

    • You must set the full-pcpus-only: true policy option so that the static CPU Manager policy only allocates full physical cores.

    • You must reserve 1 CPU for the system by setting cpu: "1" in the kubeReserved settings. This ensures that the cluster remains stable, by requiring that the system’s core functions always have access to the CPU that they need to work correctly.

  2. Run the following command to apply the changes to the KubeletConfig CR:

    $ oc apply -f <filename>.yaml
Verification

  • Inspect the kubelet configuration on a node where the change you applied the change, by running the following command and inspecting the output:

    $ oc debug node/<node_name> -- chroot /host cat /etc/kubernetes/kubelet.conf | grep -E -A 2 'cpuManagerPolicy|kubeReserved'

    Example output:

    cpuManagerPolicy: static
cpuManagerPolicyOptions:
  full-pcpus-only: true
--
kubeReserved:
  cpu: "1"