Methodology
Managed Kubernetes benchmark methodology
How managed Kubernetes cluster creation time, node join time and conformance are measured with the provider API, kube-burner and Sonobuoy.
Version 1.0 Updated:
Scope
- Managed Kubernetes clusters with a 3-node pool of the 4 vCPU shape
- Cluster creation time (API call to all nodes Ready)
- Node join time when scaling the pool from 3 to 6 nodes
- CNCF conformance (Sonobuoy certified-conformance mode)
- Control-plane responsiveness under object churn (kube-burner)
| Tool | Version |
|---|---|
| kube-burner | Recorded per run |
| Sonobuoy | Recorded per run |
Cluster configuration
The latest Kubernetes minor version offered by every provider in the comparison is used, with default networking and the default storage class. Node pools use the 4 vCPU general-purpose shape. Clusters are created through the provider’s API or Terraform provider, never the console, so timings are reproducible.
Measurements
- Cluster creation. From the create API call to
kubectl get nodesshowing 3 Ready nodes. - Node join. Pool resized to 6; time until all new nodes are Ready.
- Load balancer. Time from creating a
Serviceof typeLoadBalancerto an external address responding. - Conformance.
sonobuoy run --mode=certified-conformance; pass/fail counts recorded. - Control-plane churn. kube-burner
cluster-densityworkload; API latency p99 reported.
Repetitions and verification
Creation and join times are measured on 3 clusters created on different days. Verification requires a second operator’s result within ±15%, because control-plane provisioning varies more than VM performance.
Limitations
- Timings depend on region capacity at the time of testing.
- Optional add-ons (autoscaler, ingress) are not part of the measurement unless stated.