Skip to content

Latest commit

 

History

History
35 lines (27 loc) · 1.61 KB

dedicated_cpu_placement.md

File metadata and controls

35 lines (27 loc) · 1.61 KB
Raw

Dedicated CPU Placement

Some workloads require stronger guarantees in terms of latency and/or performance in order to operate acceptably. Virtink, relying on the Kubernetes CPU manager, is able to pin guest's vCPUs to the host's pCPUs.

Kubernetes CPU Manager

Kubernetes CPU manager is a mechanism that affects the scheduling of workloads. The default none policy explicitly enables the existing default CPU affinity scheme, providing no affinity beyond what the OS scheduler does automatically. To enable VMs with dedicated CPU resources, the policy of Kubernetes CPU manager should be set to static, which allows containers in Guaranteed pods with integer CPU requests access to exclusive CPUs on the node. For setting CPU manager policy refer to the Kubernetes documentation.

Requesting Dedicated CPU Resources

Setting spec.instance.cpu.dedicatedCPUPlacement to true in a VM spec will indicate the desire to allocate dedicated CPU resource to the VM.

Expressing the desired amount of VM's vCPUs must be done by setting both the guest topology in spec.instance.cpu (sockets and coresPerSocket) and corresponding number of vCPUs (counted as sockets * coresPerSocket) in spec.resources.[requests/limits].cpu.

Example:

apiVersion: virt.virtink.smartx.com/v1alpha1
kind: VirtualMachine
spec:
  instance:
    cpu:
      sockets: 2
      coresPerSocket: 1
      dedicatedCPUPlacement: true
    memory:
      size: 2Gi
  resources:
    requests:
      cpu: 2
      memory: 2.2Gi
    limits:
      cpu: 2
      memory: 2.2Gi