This version is under construction, please use an official release version

Configuration Maximums

Configuration Maximums defines the tested capacity limits of a single Kubermatic Virtualization (KubeV) cluster: the largest number of virtual machines, networks, firewall policies, subnets, services, and routes it can run reliably at once. Each limit was validated on a live cluster, not estimated. The pages that follow give the numbers, the factor that limits each one, and the steps to reproduce them on your own cluster with the ConfigMax benchmark tool.

These maximums describe one specific cluster configuration (the reference cluster documented below). A larger cluster will support higher figures, and a smaller cluster lower ones. They should be treated as a sizing reference rather than a fixed product limit, and they can be reproduced on any cluster using Running ConfigMax.

Pages in this section

  • Validated maximums: the per-capability maximums, together with the limiting factor for each figure.
  • Degradation: how many tenants a cluster carries before running workloads slow down, a separate measurement from raw capacity.
  • Running ConfigMax: instructions for reproducing these figures on your own cluster.
  • Engineering reference: how the numbers were measured, the per-test method cards, and the cluster tuning they depend on.

Key terms

TermDefinition
Validated maximumA count that the cluster reached and that was confirmed to remain functional, not merely accepted by the API.
Lower boundA run that reached its configured cap before the cluster exhibited any strain. The figure should be read as “at least N”; the true ceiling is higher.

Reference cluster

All figures were measured on the following cluster.

PropertyValue
Kubermatic Virtualizationv1.1.0
Worker hosts3 × 96 cores / 251 GiB RAM
Control-plane hosts3 × 12 vCPU / 24 GiB RAM
Kubernetesv1.34.3 (kubeadm)
KubeVirtv1.5.3
CNI (software-defined network)Kube-OVN 1.14.30
Operating systemUbuntu 24.04.3 LTS, kernel 6.8
Container runtimecontainerd 1.7.29
StorageLonghorn (default)

The network components were tuned for scale testing by raising the memory and CPU limits on the Kube-OVN control-plane and per-node agents. With the components left at their stock limits, the cluster reaches lower figures.