Minimum node count and rack layoutΒΆ
To qualify for Canonical Ceph design and delivery services, Ubuntu Pro and Managed Ceph, the cloud has to consist of the following components at minimum:
Three infrastructure nodes hosting the automation infrastructure, including Metal as a Service (MAAS), Juju and the observability stack, for the Juju-based options, i.e., Charmed Ceph and charm-microceph.
Six Ceph nodes for a hyperconverged architecture, or nine nodes for a disaggregated architecture.
Required rack, power and network infrastructure.
In order to ensure sufficient power supply, non-oversubscribed network and resilience against failures, we highly recommend designing the hardware layer in the following way:
Using at least three availability zones (AZs) so that one unit of each automation service and cloud control plane service would be running in a separate AZ.
Using at least three racks so that services from different AZs would be running physically separated in different racks.
Distributing hyperconverged nodes equally across racks to ensure sufficient power supply and cooling inside a single rack.
Using at least two leaf switches per AZ to ensure high availability on the network layer inside the AZ.
Using a sufficient number of spine switches (usually one per two racks) to ensure network non-oversubscription.
Using at least two managed switches for management and provisioning.
Canonical provides capacity estimates to ensure that the cloud being built has enough resources to run customer workloads. The actual number of hyperconverged node racks and switches may vary depending on the capacity requirements.