Tested scaling guidelines for VM Runtime on GDC help you plan VM deployments and evaluate cluster capacity on Distributed Cloud software only for bare metal. Because actual capacity depends on your workloads and hardware resources, these figures represent tested benchmarks rather than hard platform limits or guaranteed thresholds.
For information about quotas, limits, and scaling for Google Distributed Cloud, see Quotas and limits.
Maximum VM scaling
We have determined through extensive testing that you can run reliably as many as 500 VMs concurrently on a cluster with 50 nodes.
Resource usage
To scale up VM deployments, make sure there are enough CPU, memory, and storage resources in the cluster for both VMs and the control plane components. For standard resource requirements for cluster node machines, see Cluster node machine prerequisites.
Note the following VM Runtime on GDC resource usage for our maximum VM scaling configuration:
- Total CPU share used by all the components in the
vm-systemnamespace is approximately 1.3 at the peak usage. - Total memory consumed by all the components in the
vm-systemnamespace is approximately 10 GiB.
Known resource issue
When the guest agent is enabled,
each VM registers a separate watcher to the API server. Therefore, the number of registered watchers in kube-apiserver increases linearly with the number of the VMs. Depending on the workloads you run, you may
encounter issues related to kube-apiserver as the number of VMs increases. If
you're affected by this issue, you can configure the kube-apiserver or
disable the guest agent if
you aren't using it.