Sizing for Red Hat OpenShift Virtualization

Edit online

Red Hat® OpenShift® Virtualization (RHOCPV) is a Red Hat OpenShift feature that enables running virtual machine workloads in Red Hat OpenShift, in place of or alongside container-based workloads.

Red Hat provides a cluster sizing guide for OpenShift Virtualization that focuses on x86 and ARM workloads. This content is an extension that focuses on IBM Z® and IBM®LinuxONE workloads. Performance characteristics are different from x86 and ARM, due to the distinct overcommitment possibilities and the architectural design of the IBM Z platform.

Note: OpenShift Virtualization manages KVM guests. Use with z/VM® is not supported and tested here.

For x86 and ARM sizing see: OpenShift Virtualization - Cluster Sizing Guide (Red Hat subscriber exclusive content)

Sizing for Operator deployment

The Operator adds several components, which contribute to the overall IFL demand of the Red Hat OpenShift cluster:

D_{I F L} = 0.28 I F L s

Sizing for OpenShift Virtualization VMs

You can use the formula provided here to estimate the baseline IFL demand for operating VMs. It covers both the costs of managing a VM that uses RHOCPV and the costs of having a fully operational and idling guest operating system. The formula does not include the IFL demand from actual workloads that run inside the VMs. The formula can be used to estimate the idle state footprint, which is key for accurate capacity planning.

The sizing approach is based on capturing the IFL demand for a different number of VMs in a controlled RHOCPV environment. Whereas in theory a Red Hat OpenShift node supports up to 1500 pods, this number is an unfeasible target for the number of VMs per node. Instead, from 1 to 200 VMs per node are scaled, with a total sum of 600 VMs. The VMs are based on the following mini-sized VM configuration:

Table 1. Mini-sized VM configuration
Category	Value
CPU overcommitment ratio	10:1
Memory overcommitment ratio	1.5:1
VM operating system	Red Hat Enterprise Linux 9.6
VM - RHOCP CPU request	100 millicores
VM - RHOCP memory request	200 MiB

Each created VM has a total of one vCPU backed by 100 millicores. Whereas the overcommitment ratio per CPU core is 10:1, the effective overcommitment ratio per IFL core is 20:1 due to having SMT-2 enabled.

The following formula estimates the IFL demand for running N VM instances in an idle state:

D_{I F L} = N * 0.012 \frac{I F L s}{v m} + 0.11 I F L s

where:

$N_{}$: Number of VMs

The following table shows the estimated IFL demand for operating a specific number of idling VMs:

Table 2. IFL demand for idling VMs
Number of VMs	IFL demand [IFL cores]
10	+0.2
50	+0.7
100	+1.4
300	+3.8
600	+7.4

For example, if you plan to migrate 100 VMs to OpenShift Virtualization, you need at least 1.4 additional IFLs to manage the VMs. This does not include the IFL demand of the workload itself, which can be much higher.