Deployment Topologies
A simplyblock deployment is described by two independent decisions. The first decides where the control plane and the Simplyblock Operator run, which defines the management scope and the blast radius of a management failure. The second decides where the storage nodes run relative to the application workloads, which defines data locality, scaling, and the lifecycle coupling between storage and compute.
Two Independent Dimensions
The placement of the control plane and operator is orthogonal to the placement of the storage nodes. Any storage topology can be combined with either management topology.
| Dimension | Options | Decides |
|---|---|---|
| Control plane and operator | Local on each Kubernetes cluster, or centralized on a hub cluster | Management scope, isolation, upgrade independence, footprint on the managed clusters |
| Storage placement | Hyper-converged (HCI), disaggregated, or hybrid | Data locality, independent scaling of compute and storage, maintenance coupling |
For example, a set of small edge clusters can run hyper-converged storage while being managed from a central hub, and a large datacenter cluster can run disaggregated storage with its own local control plane.
Control Plane and Operator Placement
- Local: Every Kubernetes cluster runs its own control plane and Simplyblock Operator. This is the model that is available today.
- Hub: A hub (management) cluster runs the operator and the control plane centrally, and the managed clusters only run lightweight agents, the CSI driver, and the storage nodes. The hub model is coming soon.
The details and a comparison are in Control Plane and Operator.
Storage Placement
- Hyper-converged (HCI): Storage nodes run on the same Kubernetes workers as the applications. See HCI (Hyper-Converged).
- Disaggregated: Storage nodes run on dedicated workers, by default in the same Kubernetes cluster as the applications, which access them over NVMe-oF. See Disaggregated.
- Hybrid: Some workers host both storage and applications, while others are storage-only or compute-only. The hybrid model is a matter of node selection and does not require a different installation.
Relation to Disaster Recovery
Disaster recovery adds a hub cluster that coordinates several site clusters. The DR hub is independent of the management topology described here, but the same hub cluster can also host the centralized simplyblock control plane once that is available. The supported combinations are shown in Deployment Architectures.